'N 8@p ! /hYW2W2FW2 FW2uW2uW2ٌW2ٌW2W2W2W2W2ϜW2hϜW2W2W2GW2pGW2W2xW2W2W2kX2pkX2X2<X2{X2{X2[X2D [X2}`X2L }`X2cX2 cX2=mX2D =mX2spX2 spX2X2 X2˟X2x˟X2[X2[X2X2X2X2X23X23X2MY2 MY2 :Y2  :Y2 XY2J XY2\Y2\Y2!^Y2J!^Y2iY2ziY2 Y2 Y2Z2JZ2[2j[2[2B[2AP\2rAP\2"]2 "]24]22"} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pi b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pi b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pi  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes1152patternContiguousinvocations1} bJoperation.3exchange\"R$`b@}0 4 i   @tD<p @tD x!"#$%"&&H'Bk(ҍ)b*+,-:.2]/0R12B3)4bL5n67892:;b><a=>2?@bA+BjgC DrEFr[GʗHrI JCK~L²MNONPQRSTSUVWX"#Y"WZ2[2\B]B'^R[_R`baRb c+ dbN ep f g h i:j*]klJmnjo p,qOrqs:tʶuZvwzx Aycz:{ʨ|j} ~:3UjbM=2y2Pʄʸ Tʈʼ$Xʌ( \ ʐ   , O Jr : ʷ Z  z  B d *  J  j 3 V y  :  Z % zH  k  *  J z:]JB;w+zgFzJ~N4hʜ8lʠJ:=_Zz  */QJtږj!C:fʈZz 5*XzJz'JgB[J D x    H |    L     P     T     ! "z6 #jY ${ % & ' (:)(*ZK+m,z- ./*01J=2_3j4567 8:/9Q:Zt;<z=>?J!@CAzBҿCrD;E"xFzGH*-I*aJ*K*L*M*1N*eO*P*Q*R*5S*iT*U*V*W*9X*mY*Z*[* \*=]*q^*_*`* a0bRcud*efJghj"iDjgklm:noZp6qzYr |st*uvJw(xjKymz{|}J~z= `WR ZIZ}ZZZZMZZZZ ZQ Z Z Z Z!!ZU!Z!Z!Z!Z%"ZY"Z"Z"Z"Z)#JL#n#ʑ#Z##z# $>$*a$$J$$j$ %0%S%u%:%ʺ%Z%%z"& E&g&*&&J&& B3 U bx  Z 3 Rp   Z% a    1 e    5 i    : n ʢ    > r ʦ   B e J :  Z  z5  X z *  J  j' I l   :  Z ; z^   *  J - jP r  *  "Q z  * B ~ ڲ   N ڂ ڶ   R چ ں  " V ڊ ھ  & Z ڎ   * ^ !ʁ "Z #J $ %j  &. 'Q (t ) *: + ,Z -  .zC / f 0 1* 2 3J 4 5j5 6W 7z 8 9 :: ; <Z' =I >l ? @ AZ B- CRm D E FZ" G^ H  I  J  K 7 L k M  N  O  P ; Q o R  S  T  U ? V s W  X  Y  Z C [ w \  ]  ^  _ G ` { a b cz d  e( f*K gm hJ iڲ jj k l m= n_ o: pʤ qZ r sz t / uQ v*t w xJ y zj { ! |C! }f! ~! J! !  " I" " " 2# ># z# :# :# :$ :S$ :$ :$ :$ :#% :W% :% :% :% :'& :[& :& :& :& :+' :_' :' :' :' :/( :c( :( *( ( ( :") D) Zg) ) z)  ) ) ** 6* JY* {* j* * * + (+ :K+ m+ Z+ + z+  + , *=, _, J, , z, , )- f- -  - b. Z. . j. j/ j;/ jo/ j/ j/ j 0 j?0 js0 j0 j0 j1 jC1 jw1 j1 j1 j2 jG2 j{2 j2 j2 j3 jK3 j3 j3 Z3 3 4 j>4 `4 4 4 4 :4  5 Z05 R5 zu5  5 5 *5 5 J"6 D6 jg6 6 6 6 6 :7 67 ZY7 {7 z7 7 7 J8 E8 B8 8 :8 :9 v9 B9 9 #: W: : : : '; [; ; ; ; +< _< < < < /= c= = = = 3> g> > > !> "? # 8? $Z? %*}? &? 'J? (? )j@ *)@ +L@ ,o@ -@ .:@ /@ 0Z@ 1A 2z>A 3 aA 4A 5*A 6A 7JA 8 B 9j0B :RB ;uB <B =B >JB ?B @z"C AbC BrC CC DRD EYD FD GZD HE IBE JvE KE LE MF NFF OzF PF QF RG SJG T~G UG VG WH XNH YH ZH [H \I ]RI ^I _I `I aJ b24J c"WJ dyJ eBJ fҾJ gbJ hK i&K jIK kkK l2K m°K nRK oK prL q;L r]L s"L tL uBL vL wb M x,M yOM zrM {M |2M }M ~bM  :: y 2  *2 n ڪ 2 # W    ' [    + _    / c    3 g    z&  I k    :  Z; ] z    *  J- O jr     : A Zd  z   J 3 V *  " zN Ҋ *  ? s ڧ   C w ګ   G { گ   K  ڳ   O ڃ ڷ   B Ze J ڪ j   5 W :z ʜ Z  z  ' I *l  J  j  ; ^  :  Z   - :P r z  r. j " z  *\ * * * *, *` * * * *0 *d * * * *4 *h * * * *8 *l * * * *< !_ " # $* % &J  '. (jQ )s * + , -: .  /ZC 0e 1z 2  3 4* 5 6J5 7W 8jz 9 : ; <! =J'! >I! ?zl! @ ! A! B " CJ" D" ER" F" G<# HZx# IZ# JZ# KZ$ LZH$ MZ|$ NZ$ OZ$ PZ% QZL% RZ% SZ% TZ% UZ& VZP& WZ& XZ& YZ& ZZ ' [ZT' \Z' ]Z' ^Z' _Z$( `ZX( aJ{( bڝ( c( dZ( e) fz() g K) hm) i*) j) kJ) l) mj* n<* o_* p* q* r:* s* tZ + u.+ vzQ+ w t+ x+ y*+ z+ {J+ | , }zC, ~ f, , :, , 2'- f- *- - . 2X. . . . 0/ d/ / / 0 40 h0 0 0 1 81 l1 1 1 2 <2 p2 2 2  3 @3 t3 z3  3 3 3 "4 D4 :g4 ʉ4 Z4 4 z4  5 65 *Y5 {5 J5 5 j5 6 (6 K6 m6 :6 ʲ6 Z6 6 z7  =7 _7 :7 ڤ7 j7  8 bC8 8 Z8 8  89 bt9 9 9 : L: : : : ; P; ; ; ;  < T< < < < $= X= = = = (> \> > > :> *> ? J>? `? j? ? ? ?  @ :0@ R@ Zu@ @ z@  @ @ *"A DA JgA ډA jA A A B 6B :YB {B jB  B B :#C _C 2C C D :TD D D E 4E hE E E F 8F lF F F G L 4`L 5zL 6 L 7L 8*L 9 M :J0M ;RM <juM = M >M ?:M @M Az?N B{N CN DN EB4O FpO GO HZO IjP JjQP KzP LzP MP N!Q OUQ PQ QQ RQ S%R TYR UʍR VR WR X)S Y]S ZS [S \S ] .T ^ bT _T `T a T bU c2U dUU ewU f:U gʼU hZU iV jz$V k GV liV m*V nV oJV pV qjW r8W s[W t~W uW v:W wW xZX y*X zzMX { pX |X }:X ~X jX Y \Y Y Y Z RQZ Z Z J[ Z:[ Zn[ J[ J[ Z \ Z>\ Jr\ J\ Z\ Z] JB] Jv] Z] Z] J^ JF^ Zz^ Z^ J^ J_ ZJ_ Z~_ J_ J_ : ` +` N` Jq` ړ` j` ` ` a @a :ca ʅa Za a za  b 2b *Ub wb Jb ڼb jb c $c Gc ic :c ʮc Zc c zd 7 :z2breasonOperatorInputsvalue7invocations1 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bWuoperation.4ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue8bytes2304patternContiguousinvocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 /oF  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<5jT >s(]G|& f   bWuoperation.6ipu_stack_copy_u64%:%R)2))2)")B)Z-b1reasonLocalCopyvalue10bytes512patternContiguousinvocations4 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)")B)Z-b1reasonLocalCopyvalue10bytes256patternContiguousinvocations1  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10FF  LL)D*D44>2?2IIST^^hiss}" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 # b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1!$ bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5" bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1FF44 " "  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKllnAoAqqs+t+vvxy{{}}tt^^ӇӉHH22#* b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1$+ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2%W2%W2W2W2QX2@QX2X2 X2YX2YX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2kX2 kX2X2P X2 X2p X2Y2LY2C'?'IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$[%['')E*E,,.///11346689;x;x==?J@JBBD4E4GGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;##mm  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2uW2|uW2ٌW2ٌW2W2W2W2|W2ϜW2HϜW2W2W2AW2PAW2W2XW2W2W2kX2PkX2X2X2uX2uX2[X2$ [X2}`X2, }`X2cX2 cX2=mX2$ =mX2spX2 spX2X2| X2äX2XäX2#X2#X2eX2eX2;Y2;Y2'?'IIS T ^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7operation.6exchange :z2breasonOperatorInputsvalue8invocations1# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z9z="=zA"AzE*EbIreasonOperatorKernelvalue10kernelSpecoutputElements576input0Elements640input1Elements6272invocations1FF22    ##qq] ]""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnpgqgssuMvMxxz3{3}}rrXXˎː>>2Planned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })x2W2GW2GW2YW2dYW2W2lW2W2W2W2W2W2@W2%W2%W2W2W2QX2@QX2X2 X2YX2YX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2ӣX2 ӣX23X2D 3X2qX2dqX2Y2<Y2A;Y2A;Y2;_Y2;_Y29jY29jY2Y2Y2Z2hZ2[2l[2[2L[2O\2O\2!]2!]2;4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 4 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1h1= }  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dhP8 LpBXx@& r X > $   p&f  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL):*:44>$?$IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$I%I'')3*3,,./11346|6|88:f;f==?7@7BBD!E!GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggijlulunnpYqYssu=v=xxz!{!}}ww]]ЉЋCC))  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2FW2FW2W2dW2W2lW2W2W2W2W2ͰW2@ͰW2kW2kW2W2W2]X2@]X2X2 X2X2X2=[X2=[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2X2H X2'X2h'X2Y2DY2C*>44>(?(IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22""  ""$p%p'')\*\,,.H/H113444668 9 ;;==?h@hBBDPEPGGIPJPLLN8O8QQS T VVXY[|[|]]_d`dbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2]X2@]X2X2 X2SX2SX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2AX2P AX2X2pX2EY2LEY2DY2DY2UY2UY2`Y2`Y2Y2TY2Z2Z2[2[2m[2m[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1jjC b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1  ``8 b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1< VY*~KR&zN*"6 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>/?/IIST^^his]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnMoMqqs1t1vvxy{{}}kk݂݄OOqq  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2FW2FW2W2dW2W2lW2W2W2W2W2ǰW2@ǰW2eW2eW2W2W2]X2@]X2X2 X2X2X27[X27[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2YX2H YX2X2hX2Y2DY2C(r)`|fP8   p X @ = ( :L4 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)6*644>?IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""j j""$R%R''):*:,,."/"113 4 6~6~88:f;f==?3@3BBDEGGIJLLNNPjQjSSUNVNXXZ2[2]]_`bbdegtgtiikXlXnnp>   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2]X2@]X2X2 X2SX2SX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2X2D X2[X2d[X2=Y2<=Y2CY2CY2TY2TY2_Y2_Y2Y2DY2Z2Z2[2 [2[[2[[2O\2xO\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 5u7 b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<5jT >?sx(1Rc bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^^hhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpgqgssuKvKxxz/{/}}PP‰‹44  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2 W2l W2W2W2͛W2R͛W2ͰW2ͰW2kW2kW2W2W2iX2iX2X2zX2X2X2=[X2 =[X2bX2 bX2;lX2 ;lX2qoX2qoX2X2X2yX2yX2X2X2X2 X2gY2gY2GY2GY2 ZY2J" ZY2 eY2$ eY2Y2J%Y2Z2&Z2[2 )[2y[2*y[2GP\2,GP\2"]20"]24]22} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T5u bwBoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*Q*2reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements11521152zinput1Elementsbinvocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T5u bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T5u  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 =}}Y b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1< XF08`8h@X  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  TT)F*F44>8?8IIS$T$^^h i sese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbTT F F  88**  "#%y%y'')k*k,,.]/]113O4O668A9A;;=3>3@@B%C%EEGHJJLMO~O~QQSlTlVVXZYZ[[]H^H``b4c4eeghjjlmo{o{qqsatavvxGyG{{}-~-SSŌŎ77  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2W2TW2W2W2ӛW2ӛW2{W2({W2W2W2W2W2oX2(oX2X2X2MX2MX2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2wX2 wX2X28 X2X2XX2[X24[X2=TY2=TY2Y2,Y2Z2Z2[2[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`,  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1X b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1J8  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2FW2FW2W2dW2 W2l W2W2W2͛W2͛W2ͰW2@ͰW2kW2kW2W2W2iX2@iX2X2 X2X2X2=[X2=[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2X2 X2UX2H UX2X2hX2Y2DY2EY2EY2ZY2ZY2eY2eY2Y2LY2Z2Z2[2[2a[2a[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U5u bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U5u bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16j  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U5u  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ,,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 5u b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<5jT q>*sc(R W  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbdWeWggi=j=lln!o!qqstvtvtxxzV{V}}88ssUUƎƐ77  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2W2TW2W2W2ٛW2ٛW2{W2({W2W2W2W2W2uX2(uX2X2X2MX2MX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2 X2  X2iX2, iX2X2LX2X2$X2TY2TY2Y2Y2Z2Z2[2[2O\2O\2!]2!]2;4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ` bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ` bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 `  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1o&]B } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2ӰW2(ӰW2qW2qW2W2W2{X2({X2X2X2X2X2C[X2C[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2gX20 gX2X2PX2OX2,OX2%ZY2%ZY2Y2$Y2Z2Z2#[2#[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 9y7w  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 9y7w  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 9y7w   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v::Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1?z5  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)B*B44>2?2IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$j%j'')Z*Z,,.J/J113:4:668*9*;;=>@@B C EEGGIrJrLLN^O^QQSJTJVVX6Y6[[]"^"``bceeggipjplln\o\qqsHtHvvx4y4{{} ~  nnZZБГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2GX28 GX2X2XX2UX24UX2TY2TY2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 j. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 j. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U!  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 j.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1yD b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1;p% } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2ӰW2(ӰW2qW2qW2W2W2{X2({X2X2X2X2X2C[X2C[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2gX2 gX2X20 X2 X2P X2X2,X2ZY2ZY2Y2$Y2Z2Z2 [2 [2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 8x6v bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 8x6v bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 8x6v  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 | b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p+f!/ o  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>&?&IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqvqvssu^v^xxzF{F}}..rrZZ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2ӝX2 ӝX2X2 X2MX2, MX2X2LX2X2$X2TY2TY2Y2Y2Z2Z2[2[2cO\2cO\23!]23!]23]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 k0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 k0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 k0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ws!A b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1{ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ZDy#c } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2CW2TCW2͘W2͘W2W2W2ٰW2(ٰW2wW2wW2W2W2X2(X2X2X2X2X2I[X2I[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2CX20 CX2X2PX2X2,X2ZY2ZY2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IM@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IM@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IM@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o: b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations13s bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1BF0 p  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2%W2T%W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2ۤX2 ۤX2;X28 ;X2}X2X}X2UX24UX2%TY2%TY2Y2,Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 j" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 j" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 j"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1r~2 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations19n# } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  TT)>*>44>)?)IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''s s""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2CW2TCW2͘W2͘W2W2W2ٰW2(ٰW2wW2wW2W2W2X2(X2X2X2X2X2I[X2I[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X2-X2P-X2Y2,Y2'[Y2'[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JL> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JL> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JL>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1J N  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2+W2T+W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2͝X2 ͝X2X2 X2!X2, !X2_X2L_X2X2$X2TY2TY2Y2Y2Z2Z2 [2 [2O\2O\2!]2R!]2A4]2r} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 k$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 k$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 k$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 lht b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1y bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X Bw!a } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJoperation.6exchangezOperatorKernel107 :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2W2TW2W2W2ӛW2ӛW2߰W2(߰W2}W2}W2W2W2oX2(oX2X2X2X2X2O[X2O[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X21X2P1X2UX2,UX2][Y2][Y2 Y2$ Y2Z2Z2[2[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 .n5u  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 .n5u  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 .n5u   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Mg  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLN3?3IIS!T!^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGI[J[LLNIOIQQS7T7VVX%Y%[[]^``bcexexggifjfllnOoOqqs3t3vvxy{{}}mm߂߄QQÇÉ55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2W2TW2W2W2ӛW2ӛW2߰W2(߰W2}W2}W2W2W2oX2(oX2X2X2X2X2O[X2O[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2}X20 }X2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 -m5u  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 -m5u  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 -m5u   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,LQq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14Nm  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2 W2T W2W2W2͛W2͛W2cW2(cW2W2W2W2W2iX2(iX2X2X25X25X2[X2[X2bX2 bX2;lX2 ;lX2qoX2L qoX2X2 X2X2 X2{X2, {X2X2LX2X2$X2TY2TY2Y2Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a]^Ff b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1| bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p,h$\ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2U[X2U[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2'X2P'X2wX2,wX2\Y2\Y2Y2$Y2Z2Z2/[2/[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @6v  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @6v  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @6v   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $)I b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1_ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ٛW2ٛW2]W2(]W2W2W2W2W2uX2(uX2X2X2/X2/X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2ɤX2 ɤX2)X28 )X2kX2XkX2[X24[X2TY2TY2Y2,Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_4  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1dd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1GU8x } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2߰W2(߰W2}W2}W2W2W2uX2(uX2X2X2X2X2O[X2O[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2UX20 UX2X2PX2Y2,Y2\Y2\Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @6v bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @6v bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @6v  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 "B b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`8h(h } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ٛW2ٛW2]W2(]W2W2W2W2W2uX2(uX2X2X2/X2/X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2{X2, {X2X2LX2X2$X2?UY2?UY2Y2Y2Z2Z2[2[2O\2O\2!]2!]2M4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 VRRGg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1c*q } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>)?)IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpYqYssu=v=xxz!{!}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2[[X2[[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2}X20 }X2X2PX2[X2,[X2\Y2\Y2#Y2$#Y2Z2Z25[25[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 #c5u bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 #c5u bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 #c5u  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Jj b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1cF)@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2դX2 դX25X28 5X2wX2XwX2[X24[X2UY2UY2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T)  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1kgnHhm! b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R$v } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?Q@QBBD7E7GGIJLLNOQvQvSSU\V\XXZB[B]]_(`(bbdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2U[X2U[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2 X20  X2OX2POX2Y2,Y2g]Y2g]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 "b5u bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 "b5u bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 "b5u  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Gg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 oR5# } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2X2 X2}X2 }X2X2, X2X2LX2X2$X2UY2UY2Y2Y2Z2Z2[2[2uO\2uO\2E!]2E!]23]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 KGG b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Hl>"b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  QQ)7*744>?IISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?/@/BBDEGGIIKdLdNNPFQFSSU(V(XXZ [ ]{]{__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2/W2UW2$UW2cW2cW2W2W2'W2$'W25X25X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2CNY2| CNY2/Y2 /Y2Z2 Z2A[2 A[2kP\2 kP\2;"]2;"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdnenggiXjXllnBoBqqs,t,vvxy{{}~uu__ԇԉII33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2%W2%W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2YX2`YX2X2X2!X2!X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1w  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  aa)K*K44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44!!   ""$j%j'')T*T,,.>/>113(4(6689;;==?q@qBBD[E[GGIDJDLLN,O,QQSTVVXXZp[p]]_X`Xbbd@e@ggi(j(llnoqqssulvlxxzT{T}}<<$$ ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W25W2UW2$UW2cW2cW2W2W2'W2$'W25X25X2[X2[X2_X2$_X2X2X2cX2cX2!X2 !X2CNY2| CNY2Y2 Y2Z2 Z2/[2 /[2O\2 O\2!]2!]2w4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzK6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzK6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzK6 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)L*L44>4?4IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI8J8LLNOQQSTVwVwXXZ][]]]_C`Cbbd)e)ggijllnnphqhssuNvNxxz4{4}}ssYY̎̐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2%W2%W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2-X2`-X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  __)I*I44>1?1IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI1J1LLNOQQSSUpVpXXZV[V]]_<`v>xxz"{"}}xx\\ωϋBB(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2sW2sW2W2W2W2W2EW2EW2X2pX2m[X2m[X2`X2`X2MX2`MX2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  gg)S*S44>;?;IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG77##  "#%%'')q*q,,.]/]113I4I668595;;=>@@BCE{E{GGIcJcLLNKOKQQS3T3VVXY[[]^`w`wbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2{X2{X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>:?:IIS(T(^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGIbJbLLNPOPQQS>T>VVX,Y,[[]^``bceeggimjmllnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2sW2sW2W2W2W2W2EW2EW2X2pX2g[X2g[X2`X2`X2SX2`SX2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1"b} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1"b} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1"b } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  aa)I*I44>-?-IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55!!  } }""$e%e'')M*M,,.5/511346689;y;y==?]@]BBDAEAGGI%J%LLN O Q{Q{SSU_V_XXZC[C]]_'`'bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2-X2`-X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ZZ)D*D44>.?.IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppZZ D D  ..wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBDG?GIIS3T3^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssee W W  II;;--  "#%%'(*|*|,,.n/n113`4`668R9R;;=D>D@@B6C6EEG(H(JJLMOOQRT{T{VVXiYi[[]W^W``bCcCeeg-h-jjlmooqqsptpvvxVyV{{}<~<""xx\\ΑΓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2%W2%W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2uX2uX2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  dd)P*P44>:?:IIS"T"^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@44((  "#%%'(*~*~,,.j/j113V4V668B9B;;=.>.@@BCEEGHJzJzLLNbObQQSJTJVVX2Y2[[]^``bcevevggi^j^llnFoFqqs.t.vvxy{{}}rrZZ·ΉBB** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2a[X2a[X2_X2_X2'X2`'X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS!T!^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbddedggiJjJlln.o.qqstvvxxzc{c}}EE''yy[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2%W2%W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)H*H44>.?.IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00    "#%t%t'')\*\,,.D/D113,4,6689;;==?p@pBBDXEXGGI>J>LLN"O"QQSTVxVxXXZ\[\]]_@`@bbd$e$ggijlzlznnp^q^ssuBvBxxz&{&}} ||``҉ҋDD(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2W2W2W2W2QW2QW2X2pX2[[X2[[X2#`X2#`X2X2`X2oX2oX2X2X27NY2L 7NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1:z} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1:z} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1:z } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  __)O*O44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??//  ""$w%w'')g*g,,.W/W113G4G668797;;='>'@@BCEEGHJ}J}LLNiOiQQSUTUVVXAYA[[]-^-``bceeghj{j{llngogqqsStSvvx?y?{{}+~+yyeeۑۓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2uX2uX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1?_  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  TT)>*>44>$?$IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiUU A A  --yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2W2W2W2W2WW2WW2X2pX2[[X2[[X2)`X2)`X2qX2`qX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1;{} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1;{} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1;{ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)E*E44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$W%W'')C*C,,.///11346689;};}==?i@iBBDUEUGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77{{cc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2oO\2T oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  TT):*:44>?IISS]o^ohhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //ww] ]""$C%C''))*),,./11335h6h88:N;N==?0@0BBDEGGIIKeLeNNPGQGSSU)V)XXZ [ ]|]|__a^b^ddf@g@iik"l"nnpqsusuuuwWxWzz|9}9kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2U[X2U[X2_X2_X2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbdbebggiLjLlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2 !]24]2"} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7operation.6exchange :z2breasonOperatorInputsvalue8invocations1 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:p122prreasonvaluehhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv``ZkernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })zoutputElements"640zinput0Elements*6144zinput1Elements252488binvocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2U[X2U[X2_X2_X2!X2`!X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)@*@44>(?(IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$S%S'')=*=,,.'/'11346688:p;p==?Z@ZBBDDEDGGI,J,LLNOQQSSUkVkXXZQ[Q]]_7`7bbdeggijlvlvnnp\q\ssuBvBxxz({(}}ggډڋMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2 GW2 GW2W2 W2yW2yW2߰W2߰W2}W2}W2KW2KW2X2pX2O[X2O[X2`X2`X2X2`X2uX2uX2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations17w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations17w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations17w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>/?/IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbdSeSggi;j;lln#o#qqs t vvxxzg{g}}OOÂÄ77uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2ݎW2ݎW2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2qX2`qX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uJ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2 GW2 GW2W2 W2yW2yW2ٰW2ٰW2wW2wW2KW2KW2X2pX2I[X2I[X2`X2`X2qX2`qX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations18x} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations18x} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations18x } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  UU)=*=44>!?!IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2ݎW2ݎW2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2ǣX2`ǣX2'X2'X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tJ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2IW2IW2ٰW2ٰW2wW2wW2W2W2X2pX2I[X2I[X2_X2_X2qX2`qX2X2X2 X2 X2+NY2L +NY2Y2 Y2Z2p Z2#[2t #[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>;?;IIS'T'^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2X2X2 X2 X2+NY2T +NY2 Y2  Y23Z2x 3Z2[2| [2GP\2\ GP\2= ]2= ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  XX)D*D44>.?.IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44((  "#%%'')r*r,,.^/^113J4J668696;;=">"@@BCEEGGInJnLLNVOVQQS>T>VVX&Y&[[]^``bbdjejggiRjRllnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2IW2IW2ӰW2ӰW2qW2qW2W2W2X2pX2C[X2C[X2_X2_X2qX2`qX2X2X2 X2 X2+NY2L +NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2 W2 W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2mX2`mX2X2X2 X2 X2+NY2T +NY2Y2 Y2Z2x Z2[2| [2O\2\ O\2]2]2m2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>"?"IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI2J2LLNOQQSSUlVlXXZP[P]]_4`4bbdeggiiknlnnnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2mW2mW2ӰW2ӰW2qW2qW2?W2?W2X2pX2C[X2C[X2`X2`X2X2`X2X2X2X2X2+NY2T +NY2Y2 Y2EZ2x EZ2#[2| #[2P\2\ P\2y ]2y ]2!3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations14t} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations14t} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations14t } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1>  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)E*E44>3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqllnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2юW2юW2uW2uW2W2W2W2W2GX2pGX2[X2[X2ubX2ubX2GX2`GX2X2X2X2X2+NY2L +NY2Y2 Y2!Z2p !Z2[2t [2O\2T O\2]2]2g2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10]]  RR)<*<44>"?"IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1]]II 5 5  !!  wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbdeggiikplpnnpVqVssu'?'IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2юW2юW2uW2uW2W2W2W2W2GX2pGX2[X2[X2ubX2ubX2GX2`GX2X2X2X2X2+NY2L +NY2Y2 Y2Z2p Z2[2t [2iO\2T iO\2_]2_]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  PP)6*644>?IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2{W2|{W2ߌW2ߌW2W2W2W2|W2՜W2H՜W2W2W2GW2PGW2W2XW2W2W2qX2PqX2X2X2{X2{X2[X2$ [X2`X2, `X2cX2 cX2CmX2$ CmX2ypX2 ypX2X2| X2X2XX2AX2AX2X2X2AY2AY27?7IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1GG;; / /  ==11%%  " # %%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnOoOqqs7t7vvxy{{}~{{ccׇ׉KK33# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2W2W2W2W2QX2@QX2X2 X2SX2SX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2;X2 ;X2X2P X2X2pX2cY2LcY2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1>^c_~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1R$I  ]  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dR$vHl> dV8! l ` 4   \  0  T 9<| bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)J*J44>@@BCEEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsqtqvvxWyW{{}=~=## {{__ёѓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2uW2|uW2ߌW2ߌW2W2W2W2|W2՜W2H՜W2W2W2GW2PGW2W2XW2W2W2qX2PqX2X2X2{X2{X2[X2$ [X2`X2, `X2cX2 cX2CmX2$ CmX2ypX2 ypX2X2| X2X2XX2_X2_X2X2X2Y2Y2Cx`-0b P   p  @   b  4T bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10FF  QQ)A*A44>)?)IIS T ^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1FF66 & &  11!! y y""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpeqessuIvIxxz-{-}}ggىًKK//# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2%W2%W2W2W2QX2@QX2X2 X2YX2YX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2}X2 }X2X2P X2X2pX2cY2LcY20?0IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$_%_'')M*M,,.;/;113)4)6689;;==?a@aBBDOEOGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ~~``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2W2W2ͰW2@ͰW2kW2kW2W2W2]X2@]X2X2 X2X2X2=[X2=[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2IX2 IX2X2H X2X2hX2m Y2Dm Y2)>Y2)>Y2_Y2_Y2jY2 jY2Y2Y2Z2pZ2[2t[2y[2Ty[2O\2O\2!]2$!]2;4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @7w bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @7w bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations17k  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @7w  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ug b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1lX2r r  ` b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dlXD0tL`:Nq<  *     t r b P<|(D bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  II)3*344>?IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^JJ 6 6  ""nnX X""$B%B''),*,,,./11346u6u88:_;_==?-@-BBDEGGIJLLNNPlQlSSURVRXXZ8[8]]_`bbdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2{W2@{W2W2W2W2W2]X2@]X2X2 X2MX2MX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2CX2 CX2X2P X2X2pX2!Y2L!Y2JY2JY2;\Y2;\Y29gY29gY2Y2TY2Z2Z2[2[2y[2y[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ffkg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 I   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<I[$m6HvZ?Q  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)D*D44>4?4IIS T ^^hisjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln`o`qqsLtLvvx8y8{{}$~$XXΌΎDD  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2FW2FW2W2dW2W2lW2W2W2W2W2ͰW2@ͰW2kW2kW2W2W2]X2@]X2X2 X2X2X2=[X2=[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2ݥX2 ݥX2=X2H =X2X2hX2Y2DY23((  _    t M  g  Z )  M@j] bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  II)/*/44>?IISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$llR R""$8%8'')*,,./1w1w335]6]88:C;C==? @ B}B}DDF_G_IIK\L\NNP>Q>SSU V XXZ[]s]s__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2]X2@]X2X2 X2SX2SX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2X2P X2%X2p%X2 Y2L Y2SMY2SMY2\Y2\Y2gY2gY2Y2TY2Z2Z2[2[2g[2g[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1{' b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 7we  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<7n%\J8/f  c , u Y bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>&?&IIST^^hhrSsS} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqzqzssubvbxxzJ{J}}22ss[[ώϐCC  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ӛW2ӛW2ӰW2@ӰW2qW2qW2W2W2oX2@oX2X2 X2X2X2C[X2C[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2X2 X2X2H X21X2h1X2uY2DuY2HY2HY2ZY2ZY2eY2eY2Y2LY2Z2Z2[2[2g[2g[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T7w bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T7w bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T7w  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 cJ-m  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<cF) oR5S {F^  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)8*844>#?#IIS T ^^hhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!m m""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI3J3LLNOQQSTVwVwXXZ_[_]]_G`Gbbd/e/ggijllnnpsqsssu[v[xxzC{C}}++ooYYΎΐCC  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ӛW2ӛW2uW2(uW2W2W2W2W2oX2(oX2X2X2GX2GX2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2IX28 IX2X2XX2wY24wY2\Y2\Y2Y2,Y2Z2Z2[2[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ] bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ] bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\(  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1!%I b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/o bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Y g.  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2 W2l W2W2W2͛W2͛W2ӰW2@ӰW2qW2qW2W2W2iX2@iX2X2 X2X2X2C[X2C[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2X2 X2X2H X27X2h7X2Y2DY2JY2JY2u[Y2u[Y2sfY2sfY2Y2LY2Z2Z2ۭ[2ۭ[2U[2U[2O\2O\2o!]2o!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U7w bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U7w bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U7w  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 <|  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<<x4p,h$`\8 q * c  G  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)6*644> ? IIST^w^whhrBsB} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikelennpIqIssu-v-xxz{}}ggلن22  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2W2TW2W2W2ٛW2ٛW2uW2(uW2W2W2W2W2uX2(uX2X2X2GX2GX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X28 X2;X2X;X2Y24Y2\Y2\Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1dd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1u <  J &f } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2ٰW2(ٰW2wW2wW2W2W2{X2({X2X2X2X2X2I[X2I[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2AX2 AX2X20 X2X2PX2GY2,GY222Y2$Y2Z2Z2[2[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 5u9y  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 5u9y  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 5u9y   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 8w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ZP   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIUJULLNCOCQQS1T1VVXY[[] ^ ``bbdrerggi`j`llnIoIqqs-t-vvxy{{}}ggقلKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX25X2 5X2X28 X2X2XX2oY24oY2[Y2[Y2Y2,Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 h. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 h. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 h.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1njk b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<F n  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2ٰW2(ٰW2wW2wW2W2W2{X2({X2X2X2X2X2I[X2I[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2ץX2 ץX27X20 7X2yX2PyX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 4t8x bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 4t8x bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 4t8x  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  2 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1}  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1F  < w # c  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2AX28 AX2X2XX2Y24Y2e\Y2e\Y2Y2,Y2Z2Z2[2[2O\2O\2u!]2u!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 i0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 i0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 i0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1: b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P  Z  3 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2CW2TCW2͘W2͘W2W2W2߰W2(߰W2}W2}W2W2W2X2(X2X2X2X2X2O[X2O[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2Y2,Y2[Y2[Y2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IO@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IO@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IO@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations17w  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T\4 t  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  YY)E*E44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')s*s,,._/_113K4K668797;;=#>#@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2%W2T%W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2/X2 /X2X28 X2X2XX2_Y24_Y2[Y2[Y2Y2,Y2Z2Z2)[2)[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 h " bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 h " bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 h "  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1c__ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1b%h= }  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  ZZ)L*L44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhZZ L L  >>00""  "#%%'')q*q,,.c/c113U4U668G9G;;=9>9@@B+C+EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3qqUUǑǓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2CW2TCW2͘W2͘W2W2W2߰W2(߰W2}W2}W2W2W2X2(X2X2X2X2X2O[X2O[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X2+X2P+X2Y2,Y2/\Y2/\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JN> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JN> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JN>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 d ( l  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2AX28 AX2X2XX2Y24Y2;\Y2;\Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 i $ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 i $ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 i $  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations15 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1+ n 1 t  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2U[X2U[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2{X2 {X2X20 X2X2PX2qY2,qY2w\Y2w\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 *j7w  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 *j7w  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 *j7w   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1h7U : } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbdeg{g{iikalannpGqGssu-v-xxz{}}ll߄߆RRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ӛW2ӛW2cW2(cW2W2W2W2W2oX2(oX2X2X25X25X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2)X2 )X2X28 X2X2XX2iY24iY2\Y2\Y2#Y2,#Y2Z2Z25[25[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h;  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1XTTx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HV  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>8?8IIS$T$^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJxJxLLNdOdQQSPTPVVX?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ӛW2ӛW2cW2(cW2W2W2W2W2oX2(oX2X2X25X25X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2/X2 /X2X28 X2X2XX2Y24Y2\Y2\Y2 Y2, Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d + r 9 Y } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2[[X2[[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2IX2 IX2X20 X2X2PX2Y2,Y2C]Y2C]Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @8x  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @8x  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @8x   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1O  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@tN  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ٛW2ٛW2WW2(WW2W2W2W2W2uX2(uX2X2X2)X2)X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2}X28 }X2X2XX2Y24Y2\Y2\Y2Y2,Y2Z2Z2)[2)[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]0  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1MIIm b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1(h bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 rD X } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKlln5o5qqstvvx y {~{~}}hh݂݄RRLJlj<<&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2U[X2U[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X25X2 5X2X20 X2X2PX2EY2,EY2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @8x bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @8x bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @8x  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 )  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(  \ 6 Z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ٛW2ٛW2]W2(]W2W2W2W2W2uX2(uX2X2X2/X2/X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2ۤX2 ۤX2;X28 ;X2}X2X}X2Y24Y2m]Y2m]Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1+ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 h : ^ * } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>'?'IIS T ^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnp[q[ssuAvAxxz'{'}} ffىًLL22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2[[X2[[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2aX20 aX2X2PX2Y2,Y2 ^Y2 ^Y2Y2$Y2Z2Z2)[2)[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _7w bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _7w bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _7w  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 -M b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1}b G  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ߛW2ߛW2QW2(QW2W2W2W2W2{X2({X2X2X2#X2#X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2դX2 դX25X28 5X2wX2XwX2Y24Y2]Y2]Y2Y2,Y2Z2Z2)[2)[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R%  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1f! b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xU2  ^ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<,,  "#%z%z'')f*f,,.R/R113>4>668*9*;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2[[X2[[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2/X2 /X2X20 X2X2PX2QY2,QY2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^7w bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^7w bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^7w  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Yy b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 3  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ,  v [ +k } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2X28 X2%X2X%X2KY24KY2?^Y2?^Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S'  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1eaa9[ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 .  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1l I &   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)>*>44>$?$IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**r r""$Z%Z'')B*B,,.*/*11346688:n;n==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp>  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2UW2$UW2]W2]W2W2W2'W2$'W2/X2/X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2INY2| INY25Y2 5Y2 Z2  Z2G[2 G[2kP\2 kP\2;"]2;"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuO- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuO- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuO- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  nn)Z*Z44>D?DIIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ>>22  "&#&%%'(**,,.t/t113`4`668L9L;;=8>8@@B$C$EEGHJJLLNlOlQQSTTTVVXI?IIIS5T5^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uugg Y Y  KK==//!!  "#%%'(*~*~,,.p/p113b4b668T9T;;=F>F@@B8C8EEG*H*JJLMOOQRT}T}VVXkYk[[]Y^Y``bEcEeeg/h/jjlmooqqsrtrvvxXyX{{}>~>$$zz^^БГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2/W2UW2$UW2cW2cW2W2W2'W2$'W25X25X2[X2[X2_X2$_X2X2X2cX2cX2!X2 !X2INY2| INY2Y2 Y2Z2 Z2/[2 /[2O\2 O\2!]2!]2k4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)P*P44>6?6IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%|%|'')d*d,,.L/L1134446689;;=>@x@xBBD`E`GGIFJFLLN*O*QQSTVVXXZd[d]]_H`Hbbd,e,ggijllnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX23X2`3X2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 } bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1  } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  cc)Q*Q44>=?=IIS%T%^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uucc Q Q  ??--   ""$n%n'')\*\,,.J/J113848668&9&;;=>@@BCEyEyGGIeJeLLNOOOQQS9T9VVX#Y#[[] ^ ``bbdhehggiNjNlln2o2qqstvvxxzg{g}}II++ }}__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2mW2mW2W2W2W2W2?W2?W2X2pX2s[X2s[X2`X2`X2'X2`'X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ^^)H*H44>.?.IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##m m""$W%W'')A*A,,.+/+11346688:t;t==?Z@ZBBD@E@GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpVqVssu^  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  aa)Q*Q44>???IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11!!  "#%y%y'')i*i,,.Y/Y113I4I668999;;=)>)@@BCEEG H JJLLNkOkQQSWTWVVXCYC[[]/^/``bceeghj}j}llnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2mW2mW2W2W2W2W2?W2?W2X2pX2m[X2m[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1^} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1^} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1^ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  \\)B*B44>$?$IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##  e e""$K%K'')1*1,,./11335p6p88:V;V==?8@8BBDEGGIIKmLmNNPOQOSSU1V1XXZ[]]__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2X2X2X2X2X2CNY2CNY2Y2 Y2Z22 Z2/[2R /[2#P\2*#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" boperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r72w5:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])I*I44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$[%['')G*G,,.3/31134668 9 ;;==?m@mBBDYEYGGICJCLLN+O+QQSTVVXXZo[o]]_W`Wbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  __)I*I44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22  ~ ~""$h%h'')R*R,,.e>ggi&j&llnoqqssujvjxxzR{R}}::"" jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2uX2uX2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>5?5IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI]J]LLNGOGQQS1T1VVXY[[]^`z`zbbddedggiNjNlln8o8qqs"t"vvx y {{}}kkUUʇʉ??)) b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2-X2`-X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Rr  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>-?-IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?[@[BBDEEEGGI-J-LLNOQQSSUlVlXXZR[R]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXˉˋ>>$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2}X2`}X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ZZ)D*D44>,?,IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  l l""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI0J0LLNOQQSSUoVoXXZU[U]]_;`;bbd!e!ggijlzlznnp`q`ssuFvFxxz,{,}}kkމދQQĎĐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2W2W2W2W2W2W2QW2QW2X2pX2a[X2a[X2#`X2#`X2X2`X2uX2uX2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations16v} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations16v} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations16v } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>8?8IIS&T&^^h i s|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI`J`LLNNONQQS1?1IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==--  "#%{%{'')g*g,,.S/S113?4?668+9+;;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99ww b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2W2W2W2W2QW2QW2X2pX2a[X2a[X2#`X2#`X2wX2`wX2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations17w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations17w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations17w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  VV)@*@44>*?*IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llVV @ @  **ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLN O QQSSUkVkXXZU[U]]_?`?bbd)e)ggijllnnplqlssuPvPxxz4{4}}nnRRĎĐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2}X2`}X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)A*A44>%?%IIS T ^{^{hhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==-- u u""$]%]'')E*E,,.-/-11346688:q;q==?U@UBBD9E9GGIJLLNOQsQsSSUWVWXXZ;[;]]_`bbdeguguiikYlYnnp=q=ssu!v!xxz{}w}w[[̈́͆== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[[X2[[X2_X2_X2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>22&&  "#%%'(*|*|,,.h/h113T4T668@9@;;=,>,@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2/[2t /[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  [[)M*M44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii[[ M M  ??11##  "#%%'')r*r,,.d/d113V4V668H9H;;=:>:@@B,C,EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2U[X2U[X2_X2_X2qX2`qX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)D*D44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$p%p'')X*X,,.@/@113(4(6689;;==?l@lBBDTETGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2ӣX2`ӣX23X23X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2sW2sW2W2W2W2W2EW2EW2X2pX2U[X2U[X2`X2`X2!X2`!X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations13s} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations13s} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations13s } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>"?"IIST^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2׎W2׎W2iW2iW2W2W2W2W2;X2p;X2[X2[X2{bX2{bX2X2`X2cX2cX2X2X21NY2L 1NY2Y2 Y2Z2p Z2)[2t )[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sF b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2sW2sW2߰W2߰W2}W2}W2EW2EW2X2pX2O[X2O[X2`X2`X2wX2`wX2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2cO\2T cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations14t} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations14t} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations14t } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]k^khhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2 GW2 GW2W2 W2IW2IW2߰W2߰W2}W2}W2W2W2X2pX2O[X2O[X2_X2_X2qX2`qX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>&?&IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2 W2 W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2X2X2 X2 X21NY2T 1NY2Y2 Y29Z2x 9Z2[2| [2GP\2\ GP\2= ]2= ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations15   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>)?)IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIQJQLLN;O;QQS%T%VVXY[[]]_n`nbbdXeXggiBjBlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2IW2IW2ٰW2ٰW2wW2wW2W2W2X2pX2I[X2I[X2_X2_X2٣X2`٣X29X29X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2 W2 W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2eX2`eX2X2X2 X2 X21NY2T 1NY2Y2 Y2!Z2x !Z2[2| [2O\2\ O\2]2]2y2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844> ? IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2MX2`MX2X2X2X2X2+NY2L +NY2 Y2  Y23Z2p 3Z2[2t [2)P\2T )P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)G*G44>/?/IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11!!   ""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2gW2gW2ӰW2ӰW2qW2qW29W29W2X2pX2C[X2C[X2 `X2 `X2kX2`kX2X2X2X2X2+NY2T +NY2Y2 Y2'Z2x 'Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations11q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations11q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations18l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations11q } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)8*844>"?"IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnppqpssuTvTxxz8{8}}rrVVȎȐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2MX2`MX2X2X2X2X2+NY2L +NY2Y2 Y2Z2p Z2[2t [2O\2T O\2]2]2O2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)=*=44>!?!IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!  i i""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2{W2|{W2ߌW2ߌW2W2W2W2|W2՜W2H՜W2W2W2MW2PMW2W2XW2W2W2qX2PqX2X2X2X2X2[X2$ [X2`X2, `X2cX2 cX2CmX2$ CmX2ypX2 ypX2X2| X2դX2XդX25X25X2wX2wX2Y2Y2;Y2\;Y2[Y2[Y2fY2fY2Y28Y2Z2Z2/[2/[2[2[2P\2lP\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pm b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pm b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pm  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1)ii"[    b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d)9rn+3dxVH:t-f   X   C z  5 u  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10CC  OO)I*I44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1CC33 # #  //w w""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllnoooqqs[t[vvxGyG{{}3~3 mm# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2QX2@QX2X2 X2SX2SX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2/X2 /X2X2H X2X2hX2QY2DQY2 0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;==?^@^BBDLELGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnXoXqqs>""  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2{W2|{W2ߌW2ߌW2W2W2W2|W2՜W2H՜W2W2W2GW2PGW2W2XW2W2W2qX2PqX2X2X2{X2{X2[X2$ [X2`X2, `X2cX2 cX2CmX2$ CmX2ypX2 ypX2X2| X25X2X5X2X2X2X2X2m Y2m Y2)>Y2\)>Y2]^Y2]^Y2[iY2[iY2Y28Y2Z2Z2[2[2[2[2 P\2l P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oj b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oj b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oj  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ssx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1].  $ d b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d]:t=Q.G  hCy  T  /  e ? @ ]U0 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10BB  KK)?*?44>+?+IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1BB..    ##qq] ]""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoqqssuovoxxzW{W}}??''kk# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2W2W2W2W2QX2@QX2X2 X2SX2SX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2}X2 }X2X2H X2X2hX2cY2DcY2>`0 P   p  @   b l  4T bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)8*844>$?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$?%?''))*),,./11335r6r88:\;\==?)@)BBDEGGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2W2W2ӰW2@ӰW2qW2qW2W2W2]X2@]X2X2 X2X2X2C[X2C[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2YX2 YX2X2H X2X2hX2kY2DkY2'=Y2'=Y2[Y2[Y2fY2 fY2Y2Y2Z2pZ2[2t[2[2T[2AP\2AP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 <|9y bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 <|9y bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 <|9y  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations19y~-  \  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d9"ri+0dwVL>w0i  " [    H   : z  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  MM);*;44> ? IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGI5J5LLNOQQS T VVXY[[]]_n`nbbdXeXggiBjBlln6o6qqs t vvx y {{}}iiނބSSȇȉ==''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2W2W2{W2@{W2W2W2W2W2]X2@]X2X2 X2MX2MX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2YX2 YX2X2H X2X2hX2Y2DY2DY2DY2'[Y2'[Y2%fY2%fY2Y2LY2Z2Z2[2[2m[2m[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Aa b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 H b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<H,uX> h0x@P  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  ZZ)F*F44>2?2IIST^^his\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln@o@qqs(t(vvxy{{}}llTT ||  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2W2W2ӰW2@ӰW2qW2qW2W2W2]X2@]X2X2 X2X2X2C[X2C[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X25X2 5X2X2H X2X2hX2Y2DY2WLI;t-f   X   C   [ 5 u  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  LL)6*644>?IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGIJLzLzNNP`Q`SSUcVcXXZI[I]]_/`/bbdeggiiknlnnnp\q\ssuBvBxxz({(}}ggډڋMM33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2ǛW2ǛW2{W2@{W2W2W2W2W2cX2@cX2X2 X2MX2MX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2X2 X2CX2H CX2X2hX2Y2DY2EFY2EFY2[Y2[Y2fY2fY2Y2LY2Z2Z2ۭ[2ۭ[2U[2U[2O\2O\2u!]2u!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 #C b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 9y PU b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<9r+dPXbJ+C|E  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>$?$IIST^~^~hhrEsE} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpXqXssu)?)IIST^^hhrQsQ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33##  ""$q%q'')]*],,.I/I113545668!9!;;= > @@BBDiEiGGIQJQLLN9O9QQS!T!VVX Y [}[}]]_e`ebbdMeMggi5j5llnoqqstvyvyxxza{a}}II11ooUU  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ٛW2ٛW2uW2(uW2W2W2W2W2uX2(uX2X2X2GX2GX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2 X2  X2kX20 kX2X2PX2mX2,mX2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 #CHh b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1_ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1f-t  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ӛW2ӛW2ٰW2@ٰW2wW2wW2W2W2oX2@oX2X2 X2X2X2I[X2I[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2X2 X2GX2H GX2X2hX2Y2DY2EY2EY2[Y2[Y2fY2fY2Y2LY2Z2Z2խ[2խ[2O[2O[2O\2O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U9y bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U9y bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U9y  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  %E b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 G b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<GUc=*qO8F2 r  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)7*744>?IIST^s^shhr:s:} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$S%S'');*;,,.#/#113 4 6688:g;g==?M@MBBD1E1GGIJLLNNPkQkSSUOVOXXZ3[3]]_`bbddfmgmiikQlQnnp5q5ssuvxxzz|o}oSSńņmmސޒ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ٛW2ٛW2uW2(uW2W2W2W2W2uX2(uX2X2X2GX2GX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2_X2 _X2X20 X2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2]O\2]O\2-!]2-!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1;I V  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  LL)2*244>?IISS]i^ihhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ooU U""$;%;'')!*!,,./1z1z335`6`88:F;F==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3gg׋׍GG b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2%W2T%W2W2W2W2W2߰W2(߰W2}W2}W2W2W2X2(X2X2X2X2X2O[X2O[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X25X2 5X2X20 X2X2PX2aX2,aX23[Y23[Y2Y2$Y2Z2Z2#[2#[2kP\2kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 1q;{  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 1q;{  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 1q;{   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Hhm b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1[&q  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>9?9IIS%T%^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMOOQQSmTmVVX[Y[[[]I^I``b5c5eeghjjl m o|o|qqsbtbvvxHyH{{}.~.llތގPP‘“ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2;X2 ;X2X20 X2X2PX2aX2,aX2ZY2ZY2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 f. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 f. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 f.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Wab  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2 GW2 GW2W2LW2W2TW2W2W2ߛW2ߛW2߰W2(߰W2}W2}W2W2W2{X2({X2X2X2X2X2O[X2O[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 0p:z bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 0p:z bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 0p:z  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 n>^ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<RK  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln"o"qqstvuvuxxzW{W}}99ooQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2)X2 )X2X20 X2X2PX2X2,X2][Y2][Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 g0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 g0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 g0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 =]b b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1&k0u '  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>#?#IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikhlhnnpLqLssu0v0xxz{}}jj܄܆PPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2U[X2U[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2+X2 +X2X20 X2X2PX2SX2,SX2Q[Y2Q[Y2Y2$Y2Z2Z2)[2)[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IQ@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IQ@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IQ@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :ZZz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T!n   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)C*C44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')[*[,,.K/K113;4;668+9+;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2SX20 SX2X2PX2[X2,[X2ZY2ZY2 Y2$ Y2Z2Z2[2[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 f" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 f" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 f"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 #m(H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1H N5 u  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2CW2TCW2͘W2͘W2W2W2߰W2(߰W2}W2}W2W2W2X2(X2X2X2X2X2O[X2O[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2/X2 /X2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JP> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JP> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JP>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |@`e b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1;U"d  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$M%M'')9*9,,.%/%11346688:s;s==?_@_BBDKEKGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2MX2 MX2X20 X2X2PX2X2,X23[Y23[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 g$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 g$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 g$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1TZ  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2sX2 sX2X20 X2X2PX2UX2,UX2[Y2[Y2Y2$Y2Z2Z2#[2#[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 &f9y  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 &f9y  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 &f9y   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :Z} b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Q"s `  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln0o0qqstvvxy{y{y}}cc؂؄MM‡‰77!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X25X2 5X2X20 X2X2PX2UX2,UX2ZY2ZY2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f7  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 d b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1I[  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJvJvLLN^O^QQSFTFVVX.Y.[[]^``bbdrerggiZjZllnBoBqqs*t*vvxy{{}}nnVVʇʉ>>&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2U[X2U[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2_X2 _X2X20 X2X2PX2X2,X2/\Y2/\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 %e9y  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 %e9y  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 %e9y   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ol b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Df7  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}cc։֋II// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2GX2 GX2X20 X2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$m6 I  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)@*@44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$l%l'')T*T,,.>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2[[X2[[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2=X2P=X2SX2,SX2S\Y2S\Y2Y2$Y2Z2Z2#[2#[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @:z  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @:z  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @:z   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 4T b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\8  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ٛW2ٛW2WW2(WW2W2W2W2W2uX2(uX2X2X2)X2)X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2}X2 }X2X20 X2X2PX2[X2,[X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[,  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 _ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T(|0 p  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2[[X2[[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2wX20 wX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @:z bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @:z bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @:z  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b=Bb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pL(l } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>!?!IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ٛW2ٛW2WW2(WW2W2W2W2W2uX2(uX2X2X2)X2)X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2%X2P%X2X2,X2G\Y2G\Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 > b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P$xL  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]m^mhhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?.@.BBDEGGIIKcLcNNPEQESSU'V'XXZ [ ]z]z__a\b\ddf>g>iik l nnpqssssuuwUxUzz|7}7kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2a[X2a[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2)X2 )X2X20 X2X2PX2aX2,aX2]Y2]Y2#Y2$#Y2Z2Z25[25[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [9y bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [9y bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [9y  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ss b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1gN5 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2QW2(QW2W2W2W2W2{X2({X2X2X2#X2#X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2X20 X2%X2P%X2aX2,aX2e\Y2e\Y2Y2$Y2Z2Z2)[2)[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P!  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Xx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1_>  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  WW)A*A44>,?,IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::** v v""$`%`'')J*J,,.4/411346689;};}==?g@gBBDQEQGGI1?1IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2QW2(QW2W2W2W2W2{X2({X2X2X2#X2#X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2 X2  X2kX20 kX2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x ,,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1|[: } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>'?'IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnTTljNj::  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2UW2$UW2WW2WW2W2W2'W2$'W2)X2)X2[X2[X2_X2$_X2X2X2cX2cX2!X2 !X2INY2| INY2AY2 AY2Z2 Z2S[2 S[2P\2 P\2w"]2w"]25]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!A  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)X*X44>F?FIIS2T2^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%%'')p*p,,.`/`113P4P668@9@;;=0>0@@B C EEGHJJLLNrOrQQS^T^VVXJYJ[[]6^6``b"c"eeghjjllnpopqqs\t\vvxHyH{{}4~4  nn b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>=?=IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIeJeLLNSOSQQSATAVVX/Y/[[]^``b c eeggipjpllnYoYqqs=t=vvx!y!{{}~ww[[͇͉??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2UW2$UW2]W2]W2W2W2'W2$'W2/X2/X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2INY2| INY2)Y2 )Y2Z2 Z2;[2 ;[2/P\2 /P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rO* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rO* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rO* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$`%`'')L*L,,.8/8113$4$6689;;==?r@rBBD^E^GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX23X2`3X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ]])G*G44>1?1IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss]] G G  11zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUrVrXXZ\[\]]_F`Fbbd0e0ggijllnoqsqsssuWvWxxz;{;}}uuYYˎː== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2mW2mW2W2W2W2W2?W2?W2X2pX2s[X2s[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdkekggiUjUlln?o?qqs)t)vvxy{{}}rr\\чщFF00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2!X2`!X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1?_  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  gg)S*S44>=?=IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC77++  "#%%' ( **,,.m/m113Y4Y668E9E;;=1>1@@BCEEG H J}J}LLNeOeQQSMTMVVX5Y5[[]^``bceyeyggiajallnIoIqqs1t1vvxy{{}~uu]]чщEE-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2mW2mW2W2W2W2W2?W2?W2X2pX2s[X2s[X2`X2`X2X2`X2WX2WX2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Z} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Z} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Z } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)I*I44>1?1IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI5J5LLNOQQSTVtVtXXZZ[Z]]_@`@bbd&e&ggi j llnnpeqessuKvKxxz1{1}}ppVVɎɐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)K*K44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33##  "#%w%w'')_*_,,.G/G113/4/6689;;==?s@sBBD[E[GGIAJALLN%O%QQS T V{V{XXZ_[_]]_C`Cbbd'e'ggi j l}l}nnpaqassuEvExxz){)}} ccՉՋGG++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2UW2UW2W2W2W2W2'W2'W2X2pX2m[X2m[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)P*P44>8?8IIS T ^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD44   " # %%'')n*n,,.Z/Z113F4F668292;;=>@@BCExExGGI`J`LLNHOHQQS0T0VVXY[[]^`t`tbbd\e\ggiDjDlln,o,qqstvvxxzp{p}}XX̂̄@@%% ~~ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2 X2` X2iX2iX2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.N  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  WW)A*A44>'?'IIS T ^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llXX D D  00||f f""$P%P''):*:,,.$/$11346688:m;m==?S@SBBD9E9GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiikilinnpOqOssu5v5xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2UW2UW2W2W2W2W2'W2'W2X2pX2m[X2m[X2_X2_X2'X2`'X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ^^)F*F44>*?*IIST^^hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22 z z""$b%b'')J*J,,.2/211346689;v;v==?Z@ZBBD>E>GGI"J"LLNOQxQxSSU\V\XXZ@[@]]_$`$bbdegzgziik^l^nnpBqBssu&v&xxz { }|}|``҄҆BB$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2{O\2T {O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  WW)=*=44>?IIST^r^rhhrRsR} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22zz` `""$F%F''),*,,,./11335k6k88:Q;Q==?3@3BBDEGGIIKhLhNNPJQJSSU,V,XXZ[]]__aabaddfCgCiik%l%nnpqsxsxuuwZxZzz|<}<nnދލNN b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2yW2yW2W2W2W2W2KW2KW2X2pX2g[X2g[X2`X2`X2X2`X2X2X2X2X2=NY2L =NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations12r} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations12r} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations12r } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  bb)T*T44>D?DIIS0T0^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppbb T T  FF88**  "#%%'(*y*y,,.k/k113]4]668O9O;;=A>A@@B3C3EEG%H%JJLMOOQRTxTxVVXfYf[[]T^T``b@c@eeg*h*jjlmooqqsmtmvvxSyS{{}9~9uuYYˑ˓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  XX)B*B44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++ w w""$a%a'')K*K,,.5/51134668 9 ;~;~==?h@hBBDRERGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2W2W2W2W2W2W2QW2QW2X2pX2a[X2a[X2#`X2#`X2-X2`-X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations13s} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations13s} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations13s } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Tt  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIST^^hispsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdaeaggiGjGlln+o+qqs t v~v~xxz`{`}}BB$$vvXX b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  VV)@*@44>(?(IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //{{e e""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikmlmnnpQqQssu5v5xxz{}}ooSSƉƋ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2a[X2a[X2_X2_X2X2`X2oX2oX2X2X27NY2L 7NY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)L*L44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$t%t'')d*d,,.T/T113D4D668494;;=$>$@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[[X2[[X2_X2_X2qX2`qX2X2X2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$T%T'')@*@,,.,/,11346689;z;z==?f@fBBDRERGGI%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2sW2sW2W2W2W2W2EW2EW2X2pX2[[X2[[X2`X2`X2X2`X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2#[2t #[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1/o} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1/o} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1/o } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2X2`X2cX2cX2X2X21NY2L 1NY2 Y2  Y2Z2p Z2[2t [2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qB b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2sW2sW2W2W2W2W2EW2EW2X2pX2U[X2U[X2`X2`X2'X2`'X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations10p} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations10p} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations10p } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Tt  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2wX2`wX2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pB b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2U[X2U[X2_X2_X2wX2`wX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDlElGGITJTLLN?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?G@GBBD-E-GGIJLLNNPlQlSSURVRXXZ8[8]]_`bbdeggiikglgnnpMqMssu3v3xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2 GW2 GW2W2 W2IW2IW2߰W2߰W2}W2}W2W2W2X2pX2O[X2O[X2_X2_X2wX2`wX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44>?IIST^x^xhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2٣X2`٣X29X29X2 X2 X21NY2L 1NY2Y2 Y2 Z2p  Z2[2t [2cO\2T cO\2Y]2Y]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/+  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]n^nhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?'@'BBD E GzGzIIK\L\NNP>Q>SSU V XXZ[]s]s__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2 GW2 GW2W2 W2gW2gW2ٰW2ٰW2wW2wW29W29W2X2pX2I[X2I[X2 `X2 `X2X2`X2cX2cX2X2X2+NY2T +NY2Y2 Y2?Z2x ?Z2[2| [2qP\2\ qP\2g ]2g ]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1,l} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1,l} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1,l } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations14X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%}%}'')o*o,,.a/a113S4S668E9E;;=7>7@@B)C)EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m ooqqsotovvxUyU{{};~;!!ww[[͓͑ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2MX2`MX2X2X2X2X2+NY2L +NY2 Y2  Y23Z2p 3Z2[2t [2P\2T P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  LL)@*@44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //  k k""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`MbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%%mm b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2gW2gW2ٰW2ٰW2wW2wW29W29W2X2pX2I[X2I[X2 `X2 `X2kX2`kX2X2X2X2X2+NY2T +NY2Y2 Y2-Z2x -Z2 [2|  [2P\2\ P\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1-m} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1-m} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1-m } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)B*B44>.?.IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln/o/qqstvvxxzd{d}}FF(( zz\\ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2MX2`MX2X2X2X2X2+NY2L +NY2Y2 Y2!Z2p !Z2[2t [2O\2T O\2]2]2m2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  JJ)<*<44>$?$IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD4?4IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1@@..    (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjglln\o\qqs@t@vvx$y${{}~zz^^ЇЉBB&&# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2{W2@{W2W2W2W2W2QX2@QX2X2 X2MX2MX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2X2 X2X2H X2[X2h[X2QY2DQY2 5?5IIS!T!^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$m%m'')]*],,.M/M113=4=668-9-;;==?w@wBBDgEgGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjsllnkokqqsWtWvvxCyC{{}/~/}}iiߑߓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2{W2|{W2W2W2W2W2W2|W2ۜW2HۜW2W2W2MW2PMW2W2XW2W2W2wX2PwX2X2X2X2X2[X2$ [X2`X2, `X2cX2 cX2ImX2$ ImX2pX2 pX2X2| X2ѥX2XѥX21X21X2sX2sX2Y2Y2 3 $   pG&f  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10??  HH):*:44>&?&IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1??))    {{eeO O""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnoqtqtssuXvXxxz<{<}} vvZZ̎̐>># b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2QX2@QX2X2 X2SX2SX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2X2 X2CX2H CX2X2hX2 Y2D Y2k>Y2k>Y2-^Y2-^Y2+iY2 +iY2Y2Y2Z2pZ2[2t[2[2T[2 P\2 P\2!]2$!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 D bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 D bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 D  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  %E b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1GS m  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dGUc*q8}BL   V   ^  ` 9   . n  8x bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)=*=44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$O%O'');*;,,.'/'11346688:u;u==?B@BBBD.E.GGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2W2W2ٰW2@ٰW2wW2wW2W2W2]X2@]X2X2 X2X2X2I[X2I[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2SX2 SX2X2H X2X2hX2'Y2D'Y2;Y2;Y2\Y2\Y2gY2 gY2 Y2 Y2Z2pZ2[2t[2[2T[2AP\2AP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 8x;{ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 8x;{ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 8x;{  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1;{U    b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d;v1l'bXZN! hD/   : u 0  + f  ! \  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  VV)B*B44>?IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22&&  "#%%'')p*p,,.\/\113H4H668494;;= > @@B C EEGGIPJPLLN8O8QQS T VVX&Y&[[]^``bbdjejggiRjRllnDoDqqs,t,vvxy{{}}ppXẊ̉@@((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2ǛW2ǛW2uW2@uW2W2W2W2W2cX2@cX2X2 X2GX2GX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2_X2 _X2X2H X2X2hX2Y2DY2IHY2IHY2m]Y2m]Y2khY2khY2Y2LY2Z2Z2[2[2m[2m[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 N  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 N  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 N   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 ISX  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<I[$n7I[6 h A > bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hisZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln2o2qqstvvxy{{{{}}eeڂڄOOzz  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2W2W2ٰW2@ٰW2wW2wW2W2W2]X2@]X2X2 X2X2X2I[X2I[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2X2H X27X2h7X21 Y2D1 Y2@Y2@Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2A4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 7w;{ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 7w;{ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 7w;{  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1& a & b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d;v1l'bYO E v  = = y  4 K  / j  % ` 'g bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR):*:44>?IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$f%f'')N*N,,.6/611346689;z;z==?b@bBBDJEJGGIJLLNNPiQiSSUlVlXXZP[P]]_4`4bbdeggiiknlnnnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2ǛW2ǛW2{W2@{W2W2W2W2W2cX2@cX2X2 X2MX2MX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X21X2 1X2X2H X2X2hX29Y2D9Y2JY2JY2^Y2^Y2iY2iY2Y2LY2Z2Z2[2[2[[2[[2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a-  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 mj b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 ;{ 1  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<;v1l'bXS  s L % bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844>"?"IIST^^hhrFsF} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD?IIST^u^uhhr=s=} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$C%C'')-*-,,./11346v6v88:`;`==?H@HBBD.E.GGIJLLNNPmQmSSUSVSXXZ9[9]]_`bbdegxgxiik^l^nnpDqDssu*v*xxz{}}ii܄܆11  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ٛW2ٛW2oW2(oW2 W2 W2W2W2uX2(uX2X2X2AX2AX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2eX2 eX2X20 X2X2PX2Y2,Y2=]Y2=]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1[ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1e< j   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2 GW2 GW2W2dW2W2lW2W2W2ӛW2ӛW2߰W2@߰W2}W2}W2W2W2oX2@oX2X2 X2X2X2O[X2O[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2#X2 #X2X2H X2X2hX2Y2DY2JY2JY2\Y2\Y2gY2gY2Y2LY2Z2Z2ɭ[2ɭ[2C[2C[29O\29O\2 !]2 !]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U;{ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U;{ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U;{  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ]} b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 R(  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements1800input0Elements1800invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1<R$vHl>"t  L  ^ B bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  JJ)0*044>?IISS]g^ghhr*s*} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%mmS S""$9%9'')*,,./1x1x335^6^88:D;D==?(@(BBD E G{G{IIK]L]NNP?Q?SSU!V!XXZ[]t]t__aVbVddf8g8iiklnnpprmsmuuwOxOzz|1}1ֆֈFF&&  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ٛW2ٛW2oW2(oW2 W2 W2W2W2uX2(uX2X2X2AX2AX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2MX2 MX2X20 X2X2PX2AY2,AY2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2u!]2u!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Z  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Z  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Z   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations12  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1A  o F b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2U[X2U[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X2%X2P%X2oY2,oY2M\Y2M\Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 -m=}  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 -m=}  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 -m=}   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1f1|G  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>$?$IIS T ^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hbbd0e0ggijllnoqtqtssu\v\xxzD{D}},,tt^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2_X2 _X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2qP\2qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zI b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 d. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zI b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 d. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zI  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 d.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1:z  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HrG  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2߰W2(߰W2}W2}W2W2W2{X2({X2X2X2X2X2O[X2O[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2 X20  X2KX2PKX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ,l<| bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ,l<| bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ,l<|  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ] ( s _ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>!?!IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikflfnnpJqJssu.v.xxz{}}hhڄچNN44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2AX2 AX2X20 X2X2PX2!Y2,!Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 e0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 e0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 e0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 { b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 q F  /o } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2=X2 =X2X20 X2X2PX2oY2,oY2k\Y2k\Y2 Y2$ Y2Z2Z2[2[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IS@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IS@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IS@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Llq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1m:T / } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2_X2 _X2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2/[2/[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 d" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 d" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 d"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations19y bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1+~Q$ i } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$o%o'')_*_,,.O/O113?4?668/9/;;=>@@BCEEGGIwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2CW2TCW2͘W2͘W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2U[X2U[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X29X2P9X2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JR> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JR> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JR>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ! n ;  | } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2;X2 ;X2X20 X2X2PX2#Y2,#Y2y]Y2y]Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 e$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 e$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 e$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 w J  p < } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2CX20 CX2X2PX2}Y2,}Y2\Y2\Y2Y2$Y2Z2Z2#[2#[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 "b;{  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 "b;{  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 "b;{   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v4 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"b bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1[. B } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  ZZ)F*F44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln>o>qqs&t&vvxy{{}}jjނބRRƇƉ::"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2SX2 SX2X20 X2X2PX2sY2,sY2\Y2\Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 X  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1_0 8x } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  UU)C*C44>1?1IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd`e`ggiJjJlln4o4qqstvvxy{}{}}}gg܂܄QQƇƉ;;%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2CX20 CX2X2PX2Y2,Y2U]Y2U]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 !a;{  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 !a;{  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 !a;{   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f2 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 T ' z M  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z9z="=zA"AzE*reasonOperatorKernelvalue10kernelSpecoutputElements576input0Elements640input1Elements62722Planned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })binvocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X25X2 5X2X20 X2X2PX2 Y2, Y2I]Y2I]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 R # t E   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2a[X2a[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2eX2 eX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @<|  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @<|  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @<|   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Lll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`> z H } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ٛW2ٛW2QW2(QW2W2W2W2W2uX2(uX2X2X2#X2#X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2?X2P?X2Y2,Y2%]Y2%]Y2Y2$Y2Z2Z2)[2)[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y(  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@lB  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99 yy b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2a[X2a[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2ťX2 ťX2%X20 %X2gX2PgX2-Y2,-Y2'^Y2'^Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @<| bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @<| bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @<|  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 \| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X 6  r  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbdegsgsiikWlWnnp;q;ssuvxxz{}u}uYY˄ˆ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X2EX2PEX2Y2,Y2W^Y2W^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W;{ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W;{ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W;{  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ,1Q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18! s  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2KW2(KW2W2W2W2W2{X2({X2X2X2X2X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX25X2 5X2X20 X2X2PX2Y2,Y2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {:Uu b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1(h  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1yZ ; f } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  ]])O*O44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk]] O O  AA33%%  " # %%'')t*t,,.f/f113X4X668J9J;;=<><@@B.C.EEG H JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X25X2 5X2X20 X2X2PX23Y2,3Y2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V;{ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V;{ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V;{  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Zz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 \ E . w } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2QW2(QW2W2W2W2W2{X2({X2X2X2#X2#X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2MX20 MX2X2PX2'Y2,'Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o. b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  } ^ ? G } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>3?3IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!uuWW b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2#W2UW2$UW2QW2QW2W2W2'W2$'W2#X2#X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2INY2| INY2/Y2 /Y2Z2 Z2A[2 A[2_P\2 _P\2/"]2/"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS! b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>A?AIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIiJiLLNWOWQQSETEVVX3Y3[[]!^!``bceeggitjtlln]o]qqsAtAvvx%y%{{} ~ {{__чщCC'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)T*T44>B?BIIS.T.^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44$$  "#%|%|'')l*l,,.\/\113L4L668<9<;;=,>,@@BCEEG H JJLLNnOnQQSZTZVVXFYF[[]2^2``bceeg h jjllnlolqqsXtXvvxDyD{{}0~0~~jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2#W2UW2$UW2WW2WW2W2W2'W2$'W2)X2)X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2INY2| INY2Y2 Y2Z2 Z2/[2 /[2P\2 P\2!]2!]2}4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  __)I*I44>3?3IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uu__ I I  33||f f""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUtVtXXZ^[^]]_H`Hbbd2e2ggijllnoququssuYvYxxz={=}}!!ww[[͎͐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX23X2`3X2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Pp  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)L*L44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$^%^'')J*J,,.6/6113"4"6689;;==?p@pBBD\E\GGIFJFLLN.O.QQSTVVXXZr[r]]_Z`ZbbdBeBggi*j*llnoqqssunvnxxzV{V}}>>&&jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2gW2gW2 W2 W2W2W29W29W2X2pX2y[X2y[X2 `X2 `X2X2`X2X2X2!X2!X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  kk)W*W44>A?AIIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG;;//  "###%%'(**,,.q/q113]4]668I9I;;=5>5@@B!C!EEG H JJLLNiOiQQSQTQVVX9Y9[[]!^!``b c e}e}ggiejellnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2!X2`!X2X2X2!X2!X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1@`  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>8?8IIS"T"^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGI`J`LLNJOJQQS4T4VVXY[[]^`}`}bbdgegggiQjQlln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2gW2gW2 W2 W2W2W29W29W2X2pX2y[X2y[X2 `X2 `X2X2`X2QX2QX2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1V} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1V} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1V } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)M*M44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%y%y'')a*a,,.I/I1131416689;;=>@u@uBBD]E]GGICJCLLN'O'QQS T V}V}XXZa[a]]_E`Ebbd)e)ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])G*G44>/?/IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd$e$ggi j l}l}nnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2UW2UW2W2W2W2W2'W2'W2X2pX2s[X2s[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>+?+IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?W@WBBD=E=GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikmlmnnpSqSssu9v9xxz{}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2cX2cX2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 } bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1  } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1%E  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)L*L44>4?4IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@00  "#%~%~'')j*j,,.V/V113B4B668.9.;;=>@@BCEtEtGGI\J\LLNDODQQS,T,VVXY[[]]_p`pbbdXeXggi@j@lln(o(qqstvvxxzl{l}}TTȂȄ<<!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2[W2[W2W2W2W2W2-W2-W2X2pX2m[X2m[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2 [2t  [2{O\2T {O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  YY)?*?44>!?!IIST^t^thhrTsT} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    ||b b""$H%H'').*.,,./11335m6m88:S;S==?5@5BBDEGGIIKjLjNNPLQLSSU.V.XXZ[]]__acbcddfEgEiik'l'nnp q szszuuw\x\zz|>}> ppPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)D*D44>(?(IIS T ^~^~hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00 x x""$`%`'')H*H,,.0/011346689;t;t==?X@XBBD[>]]_"`"bbdegxgxiik\l\nnp@q@ssu$v$xxz{}z}z^^ЄІ@@"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2yW2yW2W2W2W2W2KW2KW2X2pX2m[X2m[X2`X2`X2YX2`YX2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1.n} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1.n} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1.n } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  \\)F*F44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// { {""$e%e'')O*O,,.9/9113#4#668 9 ;;==?l@lBBDVEVGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77}}gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1], b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1], b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1], b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ^^)P*P44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll^^ P P  BB44&&  " # %%'')u*u,,.g/g113Y4Y668K9K;;==>=@@B/C/EEG!H!JJLMOOQQStTtVVXbYb[[]P^P``b*?*IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUȉȋ;;!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ZZ)H*H44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llZZ H H  66$$ w w""$e%e'')S*S,,.A/A113/4/6689;;= > @@BBDpEpGGI\J\LLNFOFQQS0T0VVXY[[]^`y`ybbd_e_ggiEjElln)o)qqs t v|v|xxz^{^}}@@""ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2 W2 W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X27NY2L 7NY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f5 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2a[X2a[X2_X2_X2X2`X2wX2wX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  SS)=*=44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiSS = =  ''ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?K@KBBD5E5GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2mW2mW2W2W2W2W2?W2?W2X2pX2a[X2a[X2`X2`X2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1+k} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1+k} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1+k } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2 X2` X2iX2iX2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2mW2mW2W2W2W2W2?W2?W2X2pX2[[X2[[X2`X2`X2qX2`qX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1,l} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1,l} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1,l } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2U[X2U[X2_X2_X2-X2`-X2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1b  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>?IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2!X2!X2 X2 X21NY2L 1NY2 Y2  Y23Z2p 3Z2[2t [2#P\2T #P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xG b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>(?(IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;= > @@BBDhEhGGIPJPLLN8O8QQS T VVXY[|[|]]_d`dbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2U[X2U[X2_X2_X2wX2`wX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2cO\2T cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2AX2`AX2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2O\2T O\2]2]2I2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?L@LBBD0E0GGIJLLNNPjQjSSUNVNXXZ2[2]]_`bbdegtgtiikXlXnnpr b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1(h} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1(h} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1(h } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1'K  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  PP)<*<44>%?%IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2юW2юW2iW2iW2W2W2W2W2;X2p;X2[X2[X2ubX2ubX2SX2`SX2X2X2 X2 X2+NY2L +NY2Y2 Y2EZ2p EZ2#[2t #[2wP\2T wP\2m ]2m ]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  RR)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``RR D D  66((   ""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2 GW2 GW2W2 W2aW2aW2߰W2߰W2}W2}W23W23W2X2pX2O[X2O[X2`X2`X2kX2`kX2X2X2 X2 X2+NY2T +NY2Y2 Y2!Z2x !Z2[2| [2O\2\ O\2]2]2m2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1)i} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1)i} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1)i } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL):*:44>"?"IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2юW2юW2iW2iW2W2W2W2W2;X2p;X2[X2[X2ubX2ubX2SX2`SX2X2X2 X2 X2+NY2L +NY2Y2 Y2-Z2p -Z2 [2t  [2P\2T P\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)D*D44>0?0IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2W2W2W2|W2ۜW2HۜW2W2W2SW2PSW2W2XW2W2W2wX2PwX2X2X2X2X2%[X2$ %[X2`X2, `X2cX2 cX2ImX2$ ImX2pX2 pX2X2| X2X2XX2X2X25X25X2KY2KY2@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1@@22 $ $  66((   ""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')u*u,,.a/a113M4M668999;;=>@{@{BBDgEgGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWllnKoKqqs3t3vvxy{{}~ww__ӇӉGG//  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2W2W2W2|W2ۜW2HۜW2W2W2MW2PMW2W2XW2W2W2wX2PwX2X2X2X2X2[X2$ [X2`X2, `X2cX2 cX2ImX2$ ImX2pX2 pX2X2| X2X2XX2X2X2_X2_X2 Y2 Y2e>Y2\e>Y2`Y2`Y2kY2kY2Y28Y2Z2Z2[2[2[2[2P\2lP\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1On b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1On b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1On  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1s:  \  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dsfYL ?^23% z  k  \  M  >  /Z F7 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10??  NN)D*D44>2?2IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1??--    **}}k k""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln3o3qqstvvxxzh{h}}JJ,,bb# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2{W2@{W2W2W2W2W2QX2@QX2X2 X2MX2MX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2X2 X2X2H X2[X2h[X2Y2DY2y'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')S*S,,.;/;113#4#668 9 ;;==?F@FBBD.E.GGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II--  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2 GW2 GW2W2dW2W2lW2W2W2ǛW2ǛW2߰W2@߰W2}W2}W2W2W2cX2@cX2X2 X2X2X2O[X2O[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2X2 X2X2H X2;X2h;X23Y2D3Y2;Y2;Y2 \Y2 \Y2 gY2  gY2Y2Y2Z2pZ2)[2t)[2[2T[2}P\2}P\2M"]2$M"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 4t=} bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 4t=} bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 4t=}  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1.nn+h  %  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d.=z7\t31n+h%b\  V   P  M   G   bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  PP)@*@44>?IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``PP @ @  00   x x""$h%h'')X*X,,.H/H113848668(9(;;=>@@BCEEGGIPJPLLN?IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbd e g}g}iikclcnnpIqIssu/v/xxzz|h}hہNNTTƉƋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2 GW2 GW2W2dW2W2lW2W2W2W2W2߰W2@߰W2}W2}W2W2W2]X2@]X2X2 X2X2X2O[X2O[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2X2H X2;X2h;X2 Y2D Y2;Y2;Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2[2t[2[2T[2P\2P\2!]2$!]2}4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 3s=} bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 3s=} bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 3s=}  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1=}??   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d=z7t1 na+8h%b\  V   P  M   G  5u bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  LL)8*844>?IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$J%J'')6*6,,."/"11346688:p;p==?\@\BBDHEHGGIJLLNNPoQoSSUxVxXXZ`[`]]_H`Hbbd0e0ggijllnoq|q|ssudvdxxzL{L}}44xx``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2AW2dAW2W2lW2W2W2ǛW2ǛW2uW2@uW2W2W2W2W2cX2@cX2X2 X2GX2GX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2AX2 AX2X2H X2X2hX2Y2DY2U|[u:P)xU2  l  I  &   ` ?{X bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|}hhjjڋڍJJ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2U[X2U[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2ݥX2 ݥX2=X20 =X2X2PX2qX2,qX2)\Y2)\Y2Y2$Y2Z2Z2[2[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T=} bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T=} bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T=}  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1&f bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1%~W0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)<*<44>*?*IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$Y%Y'')G*G,,.5/5113#4#6689;;==?v@vBBDdEdGGIRJRLLN'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2߰W2(߰W2}W2}W2W2W2oX2(oX2X2X2X2X2O[X2O[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2YX2 YX2X20 X2X2PX2 Y2, Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U=} bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U=} bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U=}  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 b;N } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)7*744>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz``ӉӋFF,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ٛW2ٛW2oW2(oW2 W2 W2W2W2uX2(uX2X2X2AX2AX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X25X2 5X2X20 X2X2PX2 Y2, Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Llq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^5 2r } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2#X2 #X2X20 X2X2PX2mX2,mX2A\Y2A\Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 )i?  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 )i?  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 )i?   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2RWw b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1\ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1wR-  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;=>@@BBDjEjGGIRJRLLN:O:QQS"T"VVX Y [~[~]]_f`fbbdNeNggi6j6llnoqqstvzvzxxzb{b}}JJ22rr b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W21W2T1W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2SX2 SX2X20 X2X2PX2mX2,mX2[Y2[Y2 Y2$ Y2Z2Z2[2[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 b. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 b. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xE  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 b.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1sHn  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2U[X2U[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2_X20 _X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 (h>~ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 (h>~ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 (h>~  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 /4T b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1c>g } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^t^thhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbddfngniikRlRnnp6q6ssuvxxzz|p}pTTƄƆ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W27W2T7W2W2W2W2W2iW2(iW2W2W2 W2 W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2AX2 AX2X20 X2X2PX2Y2,Y2}\Y2}\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 c0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 c0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 c0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rGqC } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  MM)7*744>#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2qX20 qX2X2PX2SX2,SX2_\Y2_\Y2Y2$Y2Z2Z2)[2)[2qP\2qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IU@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IU@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IU@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ??_ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1< bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pM* # } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)H*H44>:?:IIS&T&^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%{%{'')m*m,,._/_113Q4Q668C9C;;=5>5@@B'C'EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m o}o}qqsctcvvxIyI{{}/~/mmߌߎQQÑÓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2]W2(]W2W2W2W2W2X2(X2X2X2/X2/X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2MX2 MX2X20 X2X2PX2gX2,gX2[Y2[Y2Y2$Y2Z2Z2#[2#[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 b" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 b" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m:  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 b"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 } b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d7 A  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  ]])I*I44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii]] Q Q  EE99--!!  "#%%'(*w*w,,.c/c113O4O668;9;;;='>'@@BCEEGGIuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2ץX2 ץX27X20 7X2yX2PyX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JT> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JT> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JT>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 n , b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1dA } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2;X2 ;X2X20 X2X2PX2X2,X2M\Y2M\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 c$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 c$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n<  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 c$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]0V  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2a[X2a[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2YX2 YX2X20 X2X2PX2aX2,aX2w\Y2w\Y2Y2$Y2Z2Z2/[2/[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^=}  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^=}  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^=}   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 rd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 iH' : } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)D*D44>4?4IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln^o^qqsJtJvvx6y6{{}"~"pp\\ґғ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ӛW2ӛW2WW2(WW2W2W2W2W2oX2(oX2X2X2)X2)X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X2X2PX2[X2,[X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0PGg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1U&w T  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?O@OBBD5E5GGIJLLNOQtQtSSUZVZXXZ@[@]]_&`&bbd e ggiikelennpKqKssu1v1xxz{}}ppVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2a[X2a[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2+X2P+X2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]=}  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]=}  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]=}   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1eD# } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ӛW2ӛW2WW2(WW2W2W2W2W2oX2(oX2X2X2)X2)X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X2+X2P+X2Y2,Y2#\Y2#\Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p' b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Hj;  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  OO)5*544>?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2g[X2g[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2MX20 MX2X2PX2;X2,;X2\Y2\Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @>~  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @>~  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @>~   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e3 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1cD%Z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ٛW2ٛW2QW2(QW2W2W2W2W2uX2(uX2X2X2#X2#X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2wX20 wX2X2PX2[X2,[X2[Y2[Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 N  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 N  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W$  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 N   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ::Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X0p  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  VV)@*@44>+?+IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99)) u u""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33{{ee b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2g[X2g[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2Y2,Y2=]Y2=]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @>~ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @>~ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @>~  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~Bbg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1gH) } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ٛW2ٛW2QW2(QW2W2W2W2W2uX2(uX2X2X2#X2#X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2SX2 SX2X20 X2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 `{ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`8hH } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>&?&IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2%W2T%W2W2W2W2W2KW2(KW2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2eX20 eX2X2PX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2/[2/[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 z$$D b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1cF) ` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2 X2  X2kX20 kX2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R=} bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R=} bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R=}  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u.3S b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jQ'g } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2W2TW2W2W2ߛW2ߛW2KW2(KW2W2W2W2W2{X2({X2X2X2X2X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2AX20 AX2X2PX2Y2,Y2I]Y2I]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Yy" b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 oR5 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  QQ);*;44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`M?MIIS9T9^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1yykk ] ]  OOAA33%%  "#%%' ( **,,.t/t113f4f668X9X;;=J>J@@B@?@IIS(T(^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvjj ^ ^  RRFF::..  ""#"%%'(**,,.p/p113\4\668H9H;;=4>4@@B C EEG H JJLLNhOhQQSPTPVVX8Y8[[] ^ ``bce|e|ggidjdllnLoLqqs4t4vvxy{{}~xx``ԇԉHH00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2W2UW2$UW2QW2QW2W2W2'W2$'W2#X2#X2[X2[X2_X2$_X2X2X2X2X2!X2 !X2INY2| INY2)Y2 )Y2Z2 Z2;[2 ;[2#P\2 #P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS'T'^^h i sysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdjejggiPjPlln4o4qqstvvxxzi{i}}KK--aa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2W2W2EW2EW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  ff)N*N44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvff V V  FF66&&  "#%z%z'')b*b,,.J/J1132426689;;=>@v@vBBD^E^GGIDJDLLN(O(QQS T V~V~XXZb[b]]_F`Fbbd*e*ggijllnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2gW2gW2W2W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2CNY2L CNY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)U*U44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%}%}'')m*m,,.]/]113M4M668=9=;;=->-@@BCEEG H JJLLNoOoQQS[T[VVXGYG[[]3^3``bceeg h jjllnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2cX2cX2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!A  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ZZ)D*D44>*?*IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo[[ G G  33  i i""$S%S'')=*=,,.'/'11346688:p;p==?V@VBBD5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$]%]'')I*I,,.5/5113!4!668 9 ;;==?o@oBBD[E[GGIEJELLN-O-QQSTVVXXZq[q]]_Y`YbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2QX2QX2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ZZ)@*@44>"?"IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  }}c c""$I%I'')/*/,,./11335n6n88:T;T==?6@6BBDEGGIIKkLkNNPMQMSSU/V/XXZ[]]__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2 W2 W2W2W2-W2-W2X2pX2y[X2y[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS#T#^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbdhehggiRjRlln.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn^^ N N  >>.. z z""$d%d'')N*N,,.8/8113"4"668 9 ;;==?k@kBBDUEUGGI>J>LLN&O&QQSTVVXXZj[j]]_R`Rbbd:e:ggi"j"lln o q~q~ssufvfxxzN{N}}66||ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2W2W2W2W2-W2-W2X2pX2s[X2s[X2_X2_X2X2`X2KX2KX2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)F*F44>.?.IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI2J2LLNOQQSSUqVqXXZW[W]]_=`=bbd#e#ggi j l|l|nnpbqbssuHvHxxz.{.}}mmSSƎƐ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  YY)C*C44>+?+IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  ~~h h""$R%R'')<*<,,.&/&11346688:o;o==?Y@YBBDCECGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggiikplpnnpTqTssu8v8xxz{}}rrVVɉɋ<<"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2sW2sW2W2W2W2W2EW2EW2X2pX2s[X2s[X2`X2`X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1*j} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1*j} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1*j } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  aa)M*M44>5?5IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqee Y Y  MMAA11  " # %%'')k*k,,.W/W113C4C668/9/;;=>@@BCEuEuGGI]J]LLNEOEQQS-T-VVXY[[]]_q`qbbdYeYggiAjAlln)o)qqstvvxxzm{m}}UUɂɄ==""{{ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  XX)F*F44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjXX F F  44""u u""$c%c'')Q*Q,,.?/?113-4-6689;;= > @@BBDnEnGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2yW2yW2W2W2W2W2KW2KW2X2pX2m[X2m[X2`X2`X2!X2`!X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1+k} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1+k} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1+k } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ff  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  [[)C*C44>'?'IIS T ^}^}hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// w w""$_%_'')G*G,,.///11346688:s;s==?W@WBBD;E;GGIJLLNOQuQuSSUYVYXXZ=[=]]_!`!bbdegwgwiik[l[nnp?q?ssu#v#xxz{}y}y]]τφ??!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  TT)>*>44>(?(IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjTT > >  ((qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLN O QQSSUiViXXZS[S]]_=`=bbd'e'ggijllnnpjqjssuNvNxxz2{2}}llމދPPސ44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2[W2[W2W2W2W2W2-W2-W2X2pX2m[X2m[X2_X2_X2X2`X2uX2uX2X2X27NY2L 7NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1>^  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  __)Q*Q44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm__ Q Q  CC55''  " # %%'')v*v,,.h/h113Z4Z668L9L;;=>>>@@B0C0EEG"H"JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations10P  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJtJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2wX2`wX2X2X2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2ǣX2`ǣX2'X2'X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI8J8LLNOQQSTVrVrXXZV[V]]_:`:bbdeggijltltnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2mW2mW2W2W2W2W2?W2?W2X2pX2a[X2a[X2`X2`X2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1'g} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1'g} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1'g } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2]W2]W2W2W2W2W2/X2p/X2[X2[X2{bX2{bX2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>?IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdegzgziik`l`nnpFqFssu,v,xxz{}}kkބކOO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2mW2mW2W2W2W2W2?W2?W2X2pX2a[X2a[X2`X2`X2eX2`eX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1(h} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1(h} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1(h } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)?*?44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2]W2]W2W2W2W2W2/X2p/X2[X2[X2{bX2{bX2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2cO\2T cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]i^ihhrIsI} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3eeՋՍEE b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2OW2OW2W2W2W2W2!W2!W2X2pX2[[X2[[X2_X2_X2X2`X2X2X2 X2 X21NY2L 1NY2 Y2  Y2Z2p Z2[2t [2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2ӣX2`ӣX23X23X2 X2 X21NY2L 1NY2Y2 Y2-Z2p -Z2 [2t  [2P\2T P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vC b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1&"  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>"?"IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI2J2LLNOQQSTVvVvXXZ^[^]]_F`Fbbd8e8ggi j llnoq|q|ssudvdxxzL{L}}44zzdd b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2[[X2[[X2_X2_X2_X2`_X2X2X2 X2 X21NY2L 1NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>"?"IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2AX2`AX2X2X2 X2 X21NY2L 1NY2Y2 Y2!Z2p !Z2[2t [2O\2T O\2]2]2g2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2U[X2U[X2`X2`X2X2`X2KX2KX2 X2 X2+NY2T +NY2 Y2  Y23Z2x 3Z2[2| [2#P\2\ #P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1$d} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1$d} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1$d } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1>  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)C*C44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW25W2 5W2юW2юW2iW2iW2W2W2W2W2;X2p;X2[X2[X2ubX2ubX2SX2`SX2X2X2 X2 X2+NY2L +NY2Y2 Y29Z2p 9Z2[2t [2AP\2T AP\27 ]27 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  LL)D*D44>2?2IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2aW2aW2߰W2߰W2}W2}W23W23W2X2pX2O[X2O[X2`X2`X2kX2`kX2X2X2 X2 X2+NY2T +NY2Y2 Y2Z2x Z2[2| [2O\2\ O\2]2]2O2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1%e} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1%e} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1%e } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>?IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW25W2 5W2юW2юW2iW2iW2W2W2W2W2;X2p;X2[X2[X2ubX2ubX2SX2`SX2X2X2 X2 X2+NY2L +NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  HH):*:44>$?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^HH 2 2  {{eeO O""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2!W2!W2W2|W2W2HW2W2W2SW2PSW2W2XW2W2W2}X2P}X2X2X2X2X2%[X2$ %[X2`X2, `X2cX2 cX2OmX2$ OmX2pX2 pX2ǝX2| ǝX2ɤX2XɤX2)X2)X2kX2kX29Y29Y2;Y2\;Y2I]Y2I]Y2GhY2GhY2Y28Y2Z2Z2[2[2[2[2P\2lP\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ps b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ps b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ps  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ( b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Mi  .  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dMg4Nh7[0U $ Z  s  B   ` - m  I bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10==  LL)@*@44>+?+IIST^^his}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1==--    //  k k""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`ObbdCeCggi+j+llnoqqssuovoxxzW{W}}??''qq# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2W2W2{W2@{W2W2W2W2W2QX2@QX2X2 X2MX2MX2[X2[X2bX2 bX2#lX2 #lX2YoX2d YoX2X2 X2 X2  X2kX2H kX2X2hX2Y2DY2;Y2;Y2m]Y2m]Y2khY2 khY2 Y2 Y2Z2pZ2[2t[2[2T[2GP\2GP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 'GGg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1?M    b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d?~=|;z9x7v5\t7 3   r  1 p / o  - m  ? bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)@*@44>.?.IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;==?X@XBBDFEFGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln=o=qqs't'vvxy{{}}ppZZχωDD..  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2!W2!W2W2|W2W2HW2W2W2SW2PSW2W2XW2W2W2}X2P}X2X2X2X2X2%[X2$ %[X2`X2, `X2cX2 cX2OmX2$ OmX2pX2 pX2ǝX2| ǝX2X2XX2X2X2/X2/X2AY2AY2&?&IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1<<((    ##ooY Y""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD%?%IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$N%N'')8*8,,."/"113 4 6688:k;k==?2@2BBDEGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpeqessuKvKxxz1{1}}ppVVɎɐ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2X2X2U[X2U[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2X2 X2UX2H UX2X2hX2!Y2D!Y2;Y2;Y2=]Y2=]Y2;hY2 ;hY2Y2Y2Z2pZ2[2t[2[2T[25P\25P\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 0p? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 0p? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 0p?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 1Q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1DN    b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dDLT\ fd,;r 8 ~ i D   P   X  $d  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  MM);*;44> ? IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGI1J1LLNOQQS T VVXY[[] ^ ``bbdpepggi^j^llnQoQqqs5t5vvxy{{}}ooSSŇʼn77  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2AW2dAW2W2lW2W2W2ǛW2ǛW2uW2@uW2W2W2W2W2cX2@cX2X2 X2GX2GX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2;X2 ;X2X2H X2X2hX2Y2DY2g,?,IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLNY2>Y2]Y2]Y2hY2 hY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2S4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 /o? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 /o? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 /o?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Oot b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E M  Y b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dE OYc(k.q4 w : @  }   @ n  E  J  Pa bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  II)3*344>>HIS T ^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__II 3 3  ||ffP P""$:%:'')$*$,,./11335m6m88:W;W==?A@ABBD+E+GGIIKiLiNNPSQSSSU`V`XXZJ[J]]_4`4bbdeggijl}l}nnpiqissuMvMxxz1{1}}kk݉݋OO33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2ǛW2ǛW2uW2@uW2W2W2W2W2cX2@cX2X2 X2GX2GX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2YX2 YX2X2H X2X2hX2 Y2D Y2>Y2>Y2'^Y2'^Y2%iY2 %iY2Y2Y2Z2pZ2[2t[2y[2Ty[2O\2O\2u!]2$u!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1])  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1T(z L  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dT(|P$xL rD h : z  U  ^ 0  0  X<| bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbdegrgriikVlVnnp:q:ssuvxxzz|Q}QÁ55<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2qX20 qX2X2PX2Y2,Y2[]Y2[]Y2Y2$Y2Z2Z2[2[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Kk b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1kD v % } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  WW)C*C44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33''  "#%%'')q*q,,.]/]113I4I668595;;=!>!@@B C EEGGIoJoLLNWOWQQS?T?VVX'Y'[[]^``bbdkekggiSjSlln;o;qqs#t#vvx y {{}}ggۂۄOOÇÉ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ٛW2ٛW2iW2(iW2W2W2W2W2uX2(uX2X2X2;X2;X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2AX2 AX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2qP\2qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 { b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations17 w  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rO , 7w } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2U[X2U[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2[X20 [X2X2PX2?Y2,?Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 6V b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 /  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1O (  Z ~ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS);*;44>#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$g%g'')O*O,,.7/711346689;{;{==?c@cBBDKEKGGI3J3LLNOQQSSUmVmXXZQ[Q]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ٛW2ٛW2oW2(oW2 W2 W2W2W2uX2(uX2X2X2AX2AX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2_X2 _X2X20 X2X2PX2% Y2,% Y2K^Y2K^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 f C  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2gX20 gX2X2PX2Y2,Y2y]Y2y]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 %eA  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 %eA  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 %eA   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ::Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rM(  @ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  KK)5*544>?IIST^v^vhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?I@IBBD/E/GGIJLLNNPnQnSSUTVTXXZ:[:]]_ ` bbdegygyiik_l_nnpEqEssu+v+xxz{}}jj݄݆PP‰‹44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W21W2T1W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2MX20 MX2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2/[2/[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 `. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 `. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 `.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  @ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1{Z 9 R } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[[X2[[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2SX20 SX2X2PX2 Y2, Y2^Y2^Y2Y2$Y2Z2Z2[2[2cO\2cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 $d@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 $d@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 $d@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $)I b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^ 9  o  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]h^hhhrJsJ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__aWbWddf9g9iiklnnpprnsnuuwPxPzz|2}2ff֋֍FF b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W27W2T7W2W2W2W2W2iW2(iW2W2W2 W2 W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2eX2 eX2X20 X2X2PX2SY2,SY2i^Y2i^Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1A  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  w V 5 1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2a[X2a[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2'X2P'X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IW@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IW@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IW@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations14t  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1yV3  [ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  PP):*:44>%?%IIS T ^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--uu__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2+W2T+W2W2W2W2W2]W2(]W2W2W2W2W2X2(X2X2X2/X2/X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2AX2 AX2X20 X2X2PX2}Y2,}Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 `" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 `" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 `"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1$d  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1#e F m } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2a[X2a[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2sX20 sX2X2PX2'Y2,'Y23^Y23^Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JV> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JV> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JV>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ==] b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1m J '   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2eX2 eX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n<  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 '  i J N } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2g[X2g[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X21X2 1X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Z?  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Z?  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Z?   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ww:ZZz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1_>  6v } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>/?/IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnKoKqqs/t/vvxy{{}}iiۂۄMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ӛW2ӛW2WW2(WW2W2W2W2W2oX2(oX2X2X2)X2)X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X25X2 5X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a-  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {5[{ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1{\ = d } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)F*F44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::**  "#%x%x'')d*d,,.P/P113<4<668(9(;;=>@@BBDpEpGGIXJXLLN@O@QQS(T(VVXY[[]]_l`lbbdTeTggi#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X21X2P1X23Y2,3Y2^Y2^Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 H bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 H bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 H  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Dd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1   ` A E } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2m[X2m[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2CX2 CX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @@  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @@  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @@   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 &FFfk b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1!a  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 kM / [ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ߛW2ߛW2QW2(QW2W2W2W2W2{X2({X2X2X2#X2#X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2X20 X2+X2P+X2Y2,Y2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 L  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 L  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 L   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p# b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1C bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1}\;  3s } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ٛW2ٛW2QW2(QW2W2W2W2W2uX2(uX2X2X2#X2#X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2/X2 /X2X20 X2X2PX2AY2,AY2W^Y2W^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 6VVv b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1y X 7   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln'o'qqs t vzvzxxz\{\}}>> ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2;X2 ;X2X20 X2X2PX2Y2,Y2W^Y2W^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1'g  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18! s  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2%W2T%W2W2W2W2W2KW2(KW2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2!X20 !X2cX2PcX2Y2,Y29^Y29^Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1,l  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 g  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$r%r'')b*b,,.R/R113B4B668292;;=">"@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2KW2(KW2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2#X2 #X2X20 X2X2PX23Y2,3Y2^Y2^Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 N  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 N  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 N   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |$DDd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 L 1  { S } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)@*@44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88%%  ""$n%n'')X*X,,.B/B113,4,6689;;=>@u@uBBD_E_GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((pp b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2 W2 W2EW2EW2W2W2W2W2X2pX2[X2[X2bX2bX23X2`3X2X2X2'X2'X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>;?;IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIcJcLLNMOMQQS7T7VVX!Y![[] ^ ``bbdjejggiTjTlln>o>qqs(t(vvxy{{}}qq[[ЇЉEE// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2W2[W2$[W2KW2KW2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2INY2| INY2Y2 Y2Z2 Z2/[2 /[2O\2 O\2!]2!]2e4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  aa)K*K44>3?3IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd"e"ggijlxlxnnp\q\ssu@v@xxz${$}}zz^^щыDD** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2 W2 W2EW2EW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)J*J44>2?2IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI6J6LLNOQQSTVuVuXXZ[[[]]_A`Abbd'e'ggi j llnnpfqfssuLvLxxz2{2}}qqWWʎʐ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2 X2` X2iX2iX2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1O} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1O} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1O } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1#C  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>?>IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttbb P P  >>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIfJfLLNTOTQQSBTBVVX0Y0[[]^``b c eeggiqjqllnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2]X2]X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)O*O44>7?7IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC33  " # %%'')m*m,,.Y/Y113E4E668191;;=>@@BCEwEwGGI_J_LLNGOGQQS/T/VVXY[[]]_s`sbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  \\)F*F44>0?0IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr\\ F F  00yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUqVqXXZ[[[]]_E`Ebbd/e/ggijllnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)G*G44>+?+IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33 { {""$c%c'')K*K,,.3/311346689;w;w==?[@[BBD?E?GGI#J#LLNOQyQySSU]V]XXZA[A]]_%`%bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  hh)T*T44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD88,,  " # %%' ( **,,.n/n113Z4Z668F9F;;=2>2@@BCEEG H J~J~LLNfOfQQSNTNVVX6Y6[[]^``bcezezggibjbllnJoJqqs2t2vvxy{{}~vv^^҇҉FF.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2 W2 W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2cX2cX2X2X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!A  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  aa)S*S44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ooaa S S  EE77))  " # %%'')x*x,,.j/j113\4\668N9N;;=@>@@@B2C2EEG$H$JJLMOOQRTwTwVVXeYe[[]S^S``b?c?eeg)h)jjlmooqqsltlvvxRyR{{}8~8ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2 W2 W2W2W2-W2-W2X2pX2y[X2y[X2_X2_X2X2`X2KX2KX2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 +  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)J*J44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%v%v'')^*^,,.F/F113.4.6689;;==?r@rBBDZEZGGI@J@LLN$O$QQSTVzVzXXZ^[^]]_B`Bbbd&e&ggi j l|l|nnp`q`ssuDvDxxz({(}} ~~bbԉԋFF** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2 W2 W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ]])K*K44>7?7IIST^^hisqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo]] K K  99'' z z""$h%h'')V*V,,.D/D113242668 9 ;;=>@@BBDsEsGGI_J_LLNIOIQQS3T3VVXY[[]^`|`|bbdbebggiHjHlln,o,qqstvvxxza{a}}CC%%wwYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2sW2sW2 W2 W2W2W2EW2EW2X2pX2y[X2y[X2`X2`X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1&f} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1&f} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1&f } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  XX)B*B44>(?(IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?T@TBBD:E:GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikjljnnpPqPssu6v6xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)K*K44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++   ""$s%s'')c*c,,.S/S113C4C668393;;=#>#@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllncocqqsOtOvvx;y;{{}'~'uuaaבד b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2sW2sW2W2W2W2W2EW2EW2X2pX2s[X2s[X2`X2`X2X2`X2{X2{X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1'g} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1'g} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1'g } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  VV)<*<44>?IIST^q^qhhrQsQ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11yy_ _""$E%E'')+*+,,./11335j6j88:P;P==?2@2BBDEGGIIKgLgNNPIQISSU+V+XXZ [ ]~]~__a`b`ddfBgBiik$l$nnpqswswuuwYxYzz|;};mm݋ݍMM b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //}}i i""$U%U'')A*A,,.-/-11346689;{;{==?g@gBBDSESGGI=J=LLN%O%QQS T VVXXZi[i]]_Q`Qbbd9e9ggi!j!lln o q}q}ssuevexxzM{M}}55yyaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2[W2[W2W2W2W2W2-W2-W2X2pX2m[X2m[X2_X2_X2SX2`SX2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  YY)C*C44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, x x""$b%b'')L*L,,.6/6113 4 668 9 ;;==?i@iBBDSESGGI/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2[W2[W2W2W2W2W2-W2-W2X2pX2m[X2m[X2_X2_X2!X2`!X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Jj  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  UU)?*?44>'?'IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)>*>44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2gW2gW2W2W2W2W29W29W2X2pX2g[X2g[X2 `X2 `X2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2׎W2׎W2]W2]W2W2W2W2W2/X2p/X2[X2[X2{bX2{bX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33qq b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2gW2gW2W2W2W2W29W29W2X2pX2g[X2g[X2 `X2 `X2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1$d} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1$d} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1$d } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2]W2]W2W2W2W2W2/X2p/X2[X2[X2{bX2{bX2X2`X2{X2{X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>?IIST^u^uhhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2OW2OW2W2W2W2W2!W2!W2X2pX2a[X2a[X2_X2_X2X2`X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2#[2t #[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>2?2IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2ߣX2`ߣX2?X2?X2X2X21NY2L 1NY2Y2 Y2EZ2p EZ2#[2t #[2wP\2T wP\2m ]2m ]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13/  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMO}O}QQSkTkVVXYYY[[]G^G``b3c3eeghjjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2OW2OW2W2W2W2W2!W2!W2X2pX2[[X2[[X2_X2_X2-X2`-X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1^~  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>$?$IIST^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2GX2`GX2X2X2X2X21NY2L 1NY2Y2 Y2-Z2p -Z2 [2t  [2P\2T P\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uC b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdVeVggi?IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdeggiiklllnnpRqRssu8v8xxz{}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW25W2 5W2ˎW2ˎW2iW2iW2W2W2W2W2;X2p;X2[X2[X2obX2obX2YX2`YX2X2X2 X2 X21NY2L 1NY2#Y2 #Y2KZ2p KZ2)[2t )[2P\2T P\2 ]2 ]2'3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  OO)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //w w""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2U[X2U[X2_X2_X2)X2`)X2X2X2 X2 X21NY2T 1NY2Y2 Y2Z2x Z2[2| [2iO\2\ iO\2_]2_]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1!a} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1!a} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1@t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1!a } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1W{  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)4*444>?IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW25W2 5W2ˎW2ˎW2iW2iW2W2W2W2W2;X2p;X2[X2[X2obX2obX2YX2`YX2X2X2 X2 X21NY2L 1NY2 Y2  Y23Z2p 3Z2[2t [2P\2T P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  KK)?*?44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##qq] ]""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2!W2!W2W2|W2W2HW2W2W2YW2PYW2W2XW2W2W2}X2P}X2X2X2X2X2+[X2$ +[X2`X2, `X2cX2 cX2OmX2$ OmX2pX2 pX2ǝX2| ǝX2gX2XgX2X2X2 X2 X2Y2Y2I1?1IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1==11 % %  ==11!!   ""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdaeaggiIjIlln1o1qqstvvxy{u{u}}]]тфEE,,# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2W2W2W2@W2W2W2W2W2WX2@WX2X2 X2SX2SX2[X2[X2bX2 bX2)lX2 )lX2_oX2d _oX2X2 X2X2 X2qX2H qX2X2hX2Y2DY2 ? IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss] ]""$G%G'')1*1,,./1l1l335V6V88:@;@==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbdeggiiknlnnnpTqTssu:v:xxz { }}yy__щыCC''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2!W2!W2W2|W2W2HW2W2W2SW2PSW2W2XW2W2W2}X2P}X2X2X2X2X2%[X2$ %[X2`X2, `X2cX2 cX2OmX2$ OmX2pX2 pX2ǝX2| ǝX2X2XX2eX2eX2X2X2Y2Y2#?#IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1<<,,    11!!  i i""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbd e ggiikclcnnpGqGssu+v+xxz{}}eeׄ׆II++ # b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2W2W2W2W2WX2@WX2X2 X2SX2SX2[X2[X2bX2 bX2)lX2 )lX2_oX2d _oX2X2 X2MX2 MX2X2H X2X2hX2Y2DY2?IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qqW W""$=%='')#*#,,. / 1W1W335=6=88:#;#==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##uuUUŐŒ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2X2X2[[X2[[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2X2 X2X2H X2_X2h_X2[ Y2D[ Y2>Y2>Y2y]Y2y]Y2whY2 whY2Y2Y2Z2pZ2)[2t)[2[2T[2qP\2qP\2A"]2$A"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ,lA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ,lA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ,lA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1YY( y  J b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d4Omn<K Z)xIk<   ^  /  O  Ho@ bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  SS)E*E44>?IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaSS E E  77))   ""$x%x'')j*j,,.\/\113N4N668@9@;;=2>2@@B$C$EEGHJkJkLLNYOYQQSGTGVVXYYY[[]G^G``b3c3eeghjjlmooqqsjtjvvxPyP{{}6~6ttXXʑʓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2GW2dGW2W2lW2W2W2ǛW2ǛW2uW2@uW2W2W2W2W2cX2@cX2X2 X2GX2GX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2SX2 SX2X2H X2X2hX2Y2DY2y&?&IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd4e4ggijllnoqxqxssu`v`xxz${$}} xxbb  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2X2X2U[X2U[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2AX2 AX2X2H X2X2hX2EY2DEY2?IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGI.J.LLNOQQSTVVXY[[]]_p`pbbdVeVggi!?!IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikjljnnpNqNssu2v2xxzz|c}cՁGGRRʼnŋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2_X2 _X2X20 X2X2PX2yX2,yX2M\Y2M\Y2Y2$Y2Z2Z2#[2#[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1'g bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1']8  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>1?1IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$i%i'')Y*Y,,.I/I113949668)9);;=>@@B C EEGGIqJqLLN]O]QQSITIVVX5Y5[[]!^!``b c eeggiojolln[o[qqsGtGvvx3y3{{}~ mmYYϑϓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ٛW2ٛW2oW2(oW2 W2 W2W2W2uX2(uX2X2X2AX2AX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2YX2 YX2X20 X2X2PX2yX2,yX2q\Y2q\Y2Y2$Y2Z2Z2[2[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1+k bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1+eB ; } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnLoLqqs0t0vvxy{{}}jj܂܄NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2[[X2[[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X2=X2P=X2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1nI$r } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)9*944>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$K%K'')7*7,,.#/#11346688:q;q==?]@]BBDIEIGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--qqYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ߛW2ߛW2uW2(uW2W2W2W2W2{X2({X2X2X2GX2GX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2YX2 YX2X20 X2X2PX2Y2,Y2 ]Y2 ]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1|Y6 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2a[X2a[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2eX2 eX2X20 X2X2PX2kX2,kX2#\Y2#\Y2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 !aC  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 !aC  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 !aC   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1sL%  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hhrtst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdZeZggiDjDlln.o.qqstvvxy{w{w}}aaւքKK55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W27W2T7W2W2W2W2W2iW2(iW2W2W2 W2 W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2eX2 eX2X20 X2X2PX2sX2,sX2\Y2\Y2Y2$Y2Z2Z2#[2#[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yG  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"b bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1"`?R } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGItJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2a[X2a[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2SX2 SX2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  `B bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  `B bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  `B  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~W0 C } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2/X2 /X2X20 X2X2PX2gX2,gX2[Y2[Y2 Y2$ Y2Z2Z2[2[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IY@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IY@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IY@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 d;}  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)A*A44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$s%s'')_*_,,.K/K113747668#9#;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33ss b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2%X2 %X2X20 X2X2PX2sX2,sX2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n<  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 XxOo b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1z[<)i } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2'X2P'X2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JX> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JX> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JX>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1i@nT } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2MX2 MX2X20 X2X2PX2Y2,Y2I]Y2I]Y2Y2$Y2Z2Z2[2[2iO\2iO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Xxx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~_@  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf;?;IIS'T'^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeWW I I  ;;--  "#%|%|'')n*n,,.`/`113R4R668D9D;;=6>6@@B(C(EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m o~o~qqsdtdvvxJyJ{{}0~0nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ٛW2ٛW2]W2(]W2W2W2W2W2uX2(uX2X2X2/X2/X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2AX2 AX2X20 X2X2PX2mX2,mX2\Y2\Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Jjj b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1qR3 ` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>*?*IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22zzdd b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2ݥX2 ݥX2=X20 =X2X2PX2Y2,Y2w\Y2w\Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UA  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UA  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UA   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y , b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T)~S 5 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ӛW2ӛW2]W2(]W2W2W2W2W2oX2(oX2X2X2/X2/X2[X2[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X21X2P1X2Y2,Y2C]Y2C]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1uV7 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2s[X2s[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2wX2 wX2X20 X2X2PX2aX2,aX2\Y2\Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @B  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @B  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @B   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`@ P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>5?5IIS!T!^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjslln_o_qqsKtKvvx7y7{{}#~#qq]]ӑӓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2aX2,aX2w\Y2w\Y2Y2$Y2Z2Z2/[2/[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 J  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 J  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 J   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 <\\| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1C bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1bA  3 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2m[X2m[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2wX2 wX2X20 X2X2PX2Y2,Y27]Y27]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @B bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @B bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @B  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 \|| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`@  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2_X2 _X2X20 X2X2PX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^= } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>&?&IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnnphqhssuLvLxxz0{0}}jj܉܋NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2AX2 AX2X20 X2X2PX2SX2,SX2I]Y2I]Y2Y2$Y2Z2Z2/[2/[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  KA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  KA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  KA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bbg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lWB } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2KW2(KW2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2/X2 /X2X20 X2X2PX2aX2,aX2\Y2\Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .NNn b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1gN5` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ``)L*L44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<00$$  "#%%'(*z*z,,.f/f113R4R668>9>;;=*>*@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2IX20 IX2X2PX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r 0 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1-nl } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2%W2T%W2W2W2W2W2QW2(QW2W2W2W2W2X2(X2X2X2#X2#X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X21X2P1X2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 L  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 L  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 L   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 6 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jQG } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2W2[W2$[W2EW2EW2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2INY2| INY2/Y2 /Y2Z2 Z2A[2 A[2MP\2 MP\2"]2"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  jj)V*V44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ::&&  "#%%'')t*t,,.`/`113L4L668898;;=">"@@B C E~E~GGIfJfLLNNONQQS6T6VVXY[[]^`z`zbbdbebggiJjJlln2o2qqstvvxy{v{v}}^^҂҄FF++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ]])G*G44>-?-IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr^^ J J  66""  l l""$V%V'')@*@,,.*/*11346688:s;s==?Y@YBBD?E?GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz^^ЉЋBB&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2W2[W2$[W2EW2EW2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2INY2| INY2Y2 Y2Z2 Z2)[2 )[2O\2 O\2!]2!]2M4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  dd)L*L44>0?0IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88$$   ""$h%h'')P*P,,.8/8113 4 6689;|;|==?`@`BBDDEDGGI(J(LLN O Q~Q~SSUbVbXXZF[F]]_*`*bbdeggiikdldnnpHqHssu,v,xxz{}}ff؄؆HH**  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2W2W2KW2KW2W2W2W2W2X2pX2[X2[X2bX2bX2X2`X2WX2WX2'X2'X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2]!]2]!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 -  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ]])C*C44>%?%IIST^x^xhhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$  f f""$L%L'')2*2,,./11335q6q88:W;W==?9@9BBDEGGIIKnLnNNPPQPSSU2V2XXZ[]]__agbgddfIgIiik+l+nnp q s~s~uuw`x`zz|B}B$$ttTTĐĒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2cX2cX2!X2!X2INY2L INY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  hh)Z*Z44>J?JIIS6T6^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvhh Z Z  LL>>00""  "#%%'(**,,.q/q113c4c668U9U;;=G>G@@B9C9EEG+H+JJLMOOQRT~T~VVXlYl[[]Z^Z``bFcFeeg0h0jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2cX2cX2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ^^)H*H44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11 } }""$g%g'')Q*Q,,.;/;113%4%6689;;==?n@nBBDXEXGGIAJALLN)O)QQSTVVXXZm[m]]_U`Ubbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdgegggiMjMlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  \\)F*F44>.?.IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  k k""$U%U'')?*?,,.)/)11346688:r;r==?\@\BBDFEFGGI.J.LLNOQQSSUmVmXXZS[S]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)R*R44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%z%z'')j*j,,.Z/Z113J4J668:9:;;=*>*@@BCEEG H JJLLNlOlQQSXTXVVXDYD[[]0^0``bceeghj~j~llnjojqqsVtVvvxByB{{}.~.||hhޑޓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S! b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>7?7IIS%T%^^h i s{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI_J_LLNMOMQQS;T;VVX)Y)[[]^``bce|e|ggijjjllnSoSqqs7t7vvxy{{}}qqUULJlj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2KX2KX2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1(  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Z%Z'')F*F,,.2/21134668 9 ;;==?l@lBBDXEXGGIBJBLLN*O*QQSTVVXXZn[n]]_V`Vbbd>e>ggi&j&llnoqqssujvjxxzR{R}}::"" ~~ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2ߣX2`ߣX2?X2?X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  WW)A*A44>+?+IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmWW A A  ++tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O QQSSUlVlXXZV[V]]_@`@bbd*e*ggijllnnpmqmssuQvQxxz5{5}}ooSSŎŐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2mW2mW2 W2 W2W2W2?W2?W2X2pX2y[X2y[X2`X2`X2X2`X2oX2oX2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1"b} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1"b} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1"b } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdeeeggiOjOlln9o9qqs#t#vvx y {{}}llVVˇˉ@@** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2ݎW2ݎW2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  aa)M*M44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==11%%  "#%%'(*{*{,,.g/g113S4S668?9?;;=+>+@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2sW2sW2 W2 W2W2W2EW2EW2X2pX2y[X2y[X2`X2`X2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)C*C44>+?+IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$V%V'')@*@,,.*/*11346688:s;s==?]@]BBDGEGGGI/J/LLNOQQSSUnVnXXZT[T]]_:`:bbd e ggijlylynnp_q_ssuEvExxz+{+}}jj݉݋PPÎÐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2ݎW2ݎW2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2٣X2`٣X29X29X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  ]])E*E44>+?+IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==--   ""$q%q'')Y*Y,,.A/A113)4)6689;;==?m@mBBDUEUGGI;J;LLNOQQSTVuVuXXZY[Y]]_=`=bbd!e!ggijlwlwnnp[q[ssu?v?xxz#{#}}yy]]ωϋAA%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2W2W2W2W2-W2-W2X2pX2s[X2s[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ^^)J*J44>2?2IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>..  "#%|%|'')h*h,,.T/T113@4@668,9,;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2kX2`kX2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>!?!IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2[W2[W2W2W2W2W2-W2-W2X2pX2s[X2s[X2_X2_X2X2`X2{X2{X2X2X2=NY2L =NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  XX)@*@44>$?$IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, t t""$\%\'')D*D,,.,/,11346688:p;p==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2gW2gW2W2W2W2W29W29W2X2pX2m[X2m[X2 `X2 `X2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2/[2t /[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n< b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--ss]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2gW2gW2W2W2W2W29W29W2X2pX2g[X2g[X2 `X2 `X2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 `} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 `} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 ` } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2׎W2׎W2cW2cW2W2W2W2W25X2p5X2[X2[X2{bX2{bX2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2%GW2%GW2ÇW2 ÇW2OW2OW2W2W2W2W2!W2!W2X2pX2g[X2g[X2_X2_X2X2`X2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>4?4IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2KX2KX2X2X21NY2L 1NY2Y2 Y29Z2p 9Z2[2t [2;P\2T ;P\21 ]21 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wE b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1@<  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLNAOAQQS/T/VVXY[[] ^ ``bbdpepggi^j^llnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2%GW2%GW2ÇW2 ÇW2OW2OW2W2W2W2W2!W2!W2X2pX2a[X2a[X2_X2_X2AX2`AX2X2X2X2X21NY2L 1NY2Y2 Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1n  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>&?&IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2GX2`GX2X2X2X2X21NY2L 1NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wG b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>?IIS T ^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQvQvSSU`V`XXZJ[J]]_4`4bbdeggijl}l}nnpiqissuMvMxxz1{1}}kk݉݋OO33 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2a[X2a[X2_X2_X2X2`X2EX2EX2 X2 X21NY21NY2Y2@Y2?Z2 ?Z2[2 [2MP\2 MP\2C ]2r C ]22]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1\} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1\} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1\ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:6  b-goperation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z 1z5"5z9"9b=reasonOperatorInputsvalue7kernelSpecoutputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7  operation.6exchange :z2  reasonOperatorInputsvalue8Planned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })binvocations10bb  PP)@*@44>,?,IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2ˎW2ˎW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2obX2obX2YX2`YX2X2X2 X2 X21NY2L 1NY2Y2 Y2?Z2p ?Z2[2t [2MP\2T MP\2C ]2C ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  UU)K*K44>5?5IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11%%  " # %%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[[X2[[X2_X2_X2YX2`YX2X2X2 X2 X21NY2L 1NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1]} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1]} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bv b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1] } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM);*;44>#?#IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2;W2 ;W2ˎW2ˎW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2obX2obX2YX2`YX2X2X2 X2 X21NY2L 1NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2'W2'W2W2|W2W2HW2W2W2YW2PYW2W2XW2W2W2X2PX2X2X2X2X2+[X2$ +[X2`X2, `X2cX2 cX2UmX2$ UmX2pX2 pX2͝X2| ͝X2X2XX2qX2qX2X2X2Y2Y2a$?$IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations177##    !!  wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}ccՉՋGG++# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2W2W2W2W2]X2@]X2X2 X2SX2SX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2X2 X2eX2H eX2X2hX2m Y2Dm Y2)>Y2)>Y2_Y2_Y2jY2 jY2 Y2 Y2Z2pZ2[2t[2[2T[2#P\2#P\2!]2$!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 C bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 C bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 C  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 "BGg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1dH8x x  ^ b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1ddH,tX< o8lJ R 8    j  P X> bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)*44>-?-IIS!T!^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$u%u'')a*a,,.M/M11346688:u;u=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbd]e]ggiEjElln-o-qqstvvxxzq{q}}YY͂̈́AA((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2'W2'W2W2|W2W2HW2W2W2YW2PYW2W2XW2W2W2X2PX2X2X2X2X2+[X2$ +[X2`X2, `X2cX2 cX2UmX2$ UmX2pX2 pX2͝X2| ͝X2X2XX2X2X2YX2YX2 Y2 Y2A>Y2\A>Y2bY2bY2mY2mY2Y28Y2Z2Z2[2[2s[2s[2iO\2liO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ot b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ot b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ot  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1h:z 8 8 x b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d(<Pdx  2 D ) V  h i z H  D0j<| bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1088  PP)6*644>?IIST^y^yhhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations188$$    ##ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?.@.BBDEGGIIKcLcNNPEQESSU'V'XXZ [ ]z]z__ahbhddfJgJiik,l,nnpqssuuwaxazz|C}C%%wwWWǐǒ# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2W2lW2W2W2W2W2W2@W2%W2%W2W2W2]X2@]X2X2 X2YX2YX2[X2[X2bX2 bX2/lX2 /lX2eoX2d eoX2X2 X2/X2 /X2X2H X2X2hX2#Y2D#Y2!?!IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$W%W'')?*?,,.'/'1t1t335\6\88:D;D==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆGG))   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2X2X2[[X2[[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2X2 X2X2H X2)X2h)X2iY2DiY2%>HIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``PP @ @  00    l l""$V%V'')@*@,,.*/*11346688:s;s==?]@]BBDGEGGGI J LLNNPhQhSSUvVvXXZ^[^]]_F`Fbbd8e8ggi j llnoq|q|ssudvdxxzL{L}}44||ff  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2ǛW2ǛW2{W2@{W2W2W2W2W2cX2@cX2X2 X2MX2MX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2SX2 SX2X2H X2X2hX2iY2DiY2%;?;IIS'T'^^hhrs} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeWW I I  ;;--  "#%|%|'')n*n,,.`/`113R4R668D9D;;=6>6@@B(C(EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m ooqqsftfvvxLyL{{} ~ eeppTTƑƓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2X2X2[[X2[[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2äX2 äX2#X2H #X2eX2heX2Y2DY2y>HIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?K@KBBD5E5GGIIKkLkNNPQQQSSU^V^XXZD[D]]_*`*bbdeggiiklllnnpPqPssu4v4xxz{}}nnTTljNj::   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2W2lW2W2W2ǛW2ǛW2W2@W2W2W2W2W2cX2@cX2X2 X2SX2SX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2;X2 ;X2X2H X2X2hX2 Y2D Y2e>Y2e>Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2w4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tty b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1CI ]  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dCI OU[a$ g ) * t  m ?  0 s  6 y  <v bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln&o&qqstvyvyxxz4{4}}ssUU  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2a[X2a[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2#X2 #X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 [{ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations13 s  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1}X3  K } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)<*<44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$Y%Y'')G*G,,.5/5113#4#6689;;==?v@vBBDdEdGGITJTLLNBOBQQS0T0VVXY[[] ^ ``bbdqeqggi_j_llnHoHqqs,t,vvxy{{}}ff؂؄JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2W2TW2W2W2ٛW2ٛW2uW2(uW2W2W2W2W2uX2(uX2X2X2GX2GX2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2AX20 AX2X2PX2{Y2,{Y2w\Y2w\Y2Y2$Y2Z2Z2)[2)[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ; b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1+k  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Nh C } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>5?5IIS!T!^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjslln_o_qqsKtKvvx7y7{{}#~#qq]]ӑӓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2W2TW2W2W2ӛW2ӛW2W2(W2W2W2W2W2oX2(oX2X2X2X2X2a[X2a[X2bX2 bX2AlX2 AlX2woX2L woX2X2 X2X2 X2X20 X2AX2PAX2 Y2, Y2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1i D  z  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)4*444> ? IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$;%;'')%*%,,./11335n6n88:X;X==?B@BBBD,E,GGIJLLNOQwQwSSUaVaXXZK[K]]_5`5bbdeggi j l~l~nnpbqbssuFvFxxz*{*}}dd։֋HH,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2W2TW2W2W2ߛW2ߛW2{W2({W2W2W2W2W2{X2({X2X2X2MX2MX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2 Y2, Y2]Y2]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 5  O   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$M%M'')9*9,,.%/%11346688:s;s==?_@_BBDKEKGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2aX20 aX2X2PX2Y2,Y2U]Y2U]Y2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]E  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]E  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]E   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2R b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1G  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`9 k  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  XX)D*D44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44((  "#%%'')r*r,,.^/^113J4J668696;;=">"@@BCEEGGIpJpLLNXOXQQS@T@VVX(Y([[]^``bbdlelggiTjTlln/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2g[X2g[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2kX2 kX2X20 X2 X2P X23Y2,3Y2]Y2]Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \D bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \D bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \D  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D  v O s } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>$?$IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpTqTssu8v8xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW27W2T7W2W2W2W2W2uW2(uW2W2W2 W2 W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2GX2 GX2X20 X2X2PX2Y2,Y2%]Y2%]Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 D  b 1  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}cc։֋II// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2Y2,Y2%]Y2%]Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I[@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I[@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I[@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }{'G b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1* j  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1CqH  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)6*644>?IIST^w^whhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdegzgziik`l`nnpFqFssu,v,xxz{}}kkބކQQÉË55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2 X20  X2OX2POX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2#[2#[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1)i  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`1 9y } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)E*E44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii]] Q Q  EE99))  "#%w%w'')c*c,,.O/O113;4;668'9';;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbdSeSggi;j;lln#o#qqs t vvxxzg{g}}OOÂÄ77ww b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2SX2PSX2 Y2, Y2]Y2]Y2Y2$Y2Z2Z2[2[2iO\2iO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JZ> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JZ> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JZ>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 v M $ @ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  LL)2*244>?IISS]i^ihhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ooU U""$;%;'')!*!,,./1z1z335`6`88:F;F==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3gg׋׍GG b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X27X2P7X2Y2,Y2I]Y2I]Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 -M b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 S $ u F   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>!?!IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1p2W21GW21GW2χW2TχW2W2\W2W2W2ٛW2ٛW2W20W2W2W2W2W2uX20uX2X2X2X2X2s[X2s[X2bX2 bX2GlX2 GlX2}oX2T }oX2X2 X2՞X2 ՞X2MX2H MX2X2 X2X2X2Y2Y2 ]Y2 ]Y2 Y2 Y2Z2lZ2[2p[2P\2PP\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 RC  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 RC  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 RC   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes1152patternContiguousinvocations1} bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1;{  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(V-  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  QQ);*;44>&?&IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqvqvssu^v^xxzF{F}}..vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ٛW2ٛW2cW2(cW2W2W2W2W2uX2(uX2X2X25X25X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2+X2P+X2Y2,Y2]Y2]Y2#Y2$#Y2Z2Z25[25[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Xx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1.n bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 b7  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  [[)M*M44>???IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii[[ M M  ??11##  "#%%'')r*r,,.d/d113V4V668H9H;;=:>:@@B,C,EEGHJJLMOOQQSsTsVVXaYa[[]O^O``b;c;eeg%h%jjlmooqqshthvvxNyN{{}4~4rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2aX2 aX2X20 X2X2PX2-Y2,-Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 QC  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 QC  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 QC   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 [ 2 !a } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>#?#IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikhlhnnpLqLssu0v0xxz{}}jj܄܆PPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2W2TW2W2W2ٛW2ٛW2cW2(cW2W2W2W2W2uX2(uX2X2X25X25X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2wX20 wX2X2PX2 Y2, Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 D bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 D bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 D  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ??_ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 a 6 #c } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  UU)C*C44>1?1IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2W2TW2W2W2ߛW2ߛW2 W2( W2W2W2W2W2{X2({X2X2X2X2X2y[X2y[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2 X20  X2KX2PKX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @D  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @D  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @D   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1bD & P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnLoLqqs0t0vvxy{{}}jj܂܄NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ߛW2ߛW2WW2(WW2W2W2W2W2{X2({X2X2X2)X2)X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2X20 X2+X2P+X2Y2,Y2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 H  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 H  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 H   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1>~  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pP 0 T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)I*I44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllncocqqsOtOvvx;y;{{}'~'uuaaבד b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2s[X2s[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2wX2 wX2X20 X2X2PX2EY2,EY2{^Y2{^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @D bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @D bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @D  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 &  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  j L . 2 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2W2TW2W2W2ߛW2ߛW2]W2(]W2W2W2W2W2{X2({X2X2X2/X2/X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2 X2  X2kX20 kX2X2PX2?Y2,?Y2{^Y2{^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y --M b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  p P 0 4 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)?*?44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33ww__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2IX20 IX2X2PX2Y2,Y2^Y2^Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 GC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 GC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 GC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ppp * b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1YF3 W } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGItJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2%W2T%W2W2W2W2W2QW2(QW2W2W2W2W2X2(X2X2X2#X2#X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2_X2 _X2X20 X2X2PX2Y2,Y2^Y2^Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ]} b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hS> ) Z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>3?3IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbdbebggiLjLlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2mX2 mX2X20 X2X2PX2WY2,WY2M_Y2M_Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 FC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 FC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 FC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 2  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 z g T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)@*@44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$l%l'')T*T,,.>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2%W2T%W2W2W2W2W2WW2(WW2W2W2W2W2X2(X2X2X2)X2)X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2_X20 _X2X2PX2+ Y2,+ Y2M_Y2M_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 J  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 J  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 J   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 n; b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  j U } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)>*>44>(?(IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI,J,LLNOQQSSUkVkXXZQ[Q]]_7`7bbdeggijlvlvnnp\q\ssuBvBxxz({(}}ggډڋMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2[W2$[W2?W2?W2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2ONY2| ONY2Y2 Y2Z2 Z2/[2 /[2O\2 O\2!]2!]2k4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  aa)K*K44>1?1IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&p p""$Z%Z'')D*D,,../.11346689;w;w==?]@]BBDCECGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~bbԉԋFF** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  ff)R*R44>:?:IIS"T"^^h i s|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvjj ^ ^  RRFF66""  "#%%'')p*p,,.\/\113H4H668494;;=>@@BCEzEzGGIbJbLLNJOJQQS2T2VVXY[[]^`v`vbbd^e^ggiFjFlln.o.qqstvvxxzr{r}}ZZ΂΄BB''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2W2[W2$[W2EW2EW2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2WX2WX2'X2 'X2ONY2| ONY2 Y2  Y2Z2 Z2[2 [2O\2 O\2]!]2]!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  __)E*E44>'?'IIS T ^z^zhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&h h""$N%N'')4*4,,./11346s6s88:Y;Y==?;@;BBDEGGIIKpLpNNPRQRSSU4V4XXZ[]]__aibiddfKgKiik-l-nnpqssuuwbxbzz|D}D&&vvVVƐƒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)J*J44>.?.IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvff V V  FF66""  ~ ~""$f%f'')N*N,,.6/611346689;z;z==?^@^BBDBEBGGI&J&LLN O Q|Q|SSU`V`XXZD[D]]_(`(bbd e g~g~iikblbnnpFqFssu*v*xxz{}}ddքֆFF((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2[W2[W2!W2!W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  bb)L*L44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55""   ""$k%k'')U*U,,.?/?113)4)6689;;==?r@rBBD\E\GGIEJELLN-O-QQSTVVXXZq[q]]_Y`YbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%%mm b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2cX2cX2!X2!X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  dd)V*V44>F?FIIS2T2^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrdd V V  HH::,,  "#%%'(*{*{,,.m/m113_4_668Q9Q;;=C>C@@B5C5EEG'H'JJLMOOQRTzTzVVXhYh[[]V^V``bBcBeeg,h,jjlmooqqsotovvxUyU{{};~;!!ww[[͓͑ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1F} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1F} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1F } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ^^)H*H44>0?0IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##m m""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI0J0LLNOQQSSUoVoXXZU[U]]_;`;bbdeggijlulunnpYqYssu=v=xxz!{!}}ww[[Ή΋AA'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ``)N*N44>:?:IIS"T"^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr`` N N  <<** } }""$k%k'')Y*Y,,.G/G113545668#9#;;=>@@BBDvEvGGIbJbLLNLOLQQS6T6VVX Y [[] ^ ``bbdeeeggiKjKlln/o/qqstvvxxzd{d}}FF(( zz\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>;?;IIS)T)^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIcJcLLNQOQQQS?T?VVX-Y-[[]^``b c eegginjnllnWoWqqs;t;vvxy{{}~uuYYˇˉ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U% b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ^^)N*N44>>..  ""$v%v'')f*f,,.V/V113F4F668696;;=&>&@@BCEEGHJ|J|LLNhOhQQSTTTVVX@Y@[[],^,``bceeghjzjzllnfofqqsRtRvvx>y>{{}*~*xxddڑړ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  YY)C*C44>-?-IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ooYY C C  --vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLNOQQSSUnVnXXZX[X]]_B`Bbbd,e,ggijllnnpoqossuSvSxxz7{7}}qqUUǎǐ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2EX2EX2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1"  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ZZ)F*F44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  l l""$X%X'')D*D,,.0/011346689;~;~==?j@jBBDVEVGGI@J@LLN(O(QQSTVVXXZl[l]]_T`Tbbd;?;IIS#T#^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqee Y Y  MMAA55))  "#%%' ( **,,.k/k113W4W668C9C;;=/>/@@BCEEGHJ{J{LLNcOcQQSKTKVVX3Y3[[]^``bcewewggi_j_llnGoGqqs/t/vvxy{{}}ss[[χωCC++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2ݎW2ݎW2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2oX2oX2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^. b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  XX)F*F44>2?2IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjXX F F  44""u u""$c%c'')Q*Q,,.?/?113-4-6689;;= > @@BBDnEnGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKlln5o5qqstvvx y {~{~}}hh݂݄RRLJlj<<&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2mW2mW2W2W2W2W2?W2?W2X2pX2[X2[X2`X2`X2ߣX2`ߣX2?X2?X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  __)G*G44>-?-IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??//  ""$s%s'')[*[,,.C/C113+4+6689;;==?o@oBBDWEWGGI=J=LLN!O!QQSTVwVwXXZ[[[]]_?`?bbd#e#ggijlylynnp]q]ssuAvAxxz%{%}} {{__щыCC'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2ݎW2ݎW2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  WW)A*A44>)?)IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //}}i i""$T%T'')>*>,,.(/(11346688:q;q==?[@[BBDEEEGGI-J-LLNOQQSSUlVlXXZR[R]]_8`8bbdeggijlwlwnnp]q]ssuCvCxxz){)}}hhۉۋNN44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2 W2 W2W2W2-W2-W2X2pX2y[X2y[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  UU)?*?44>%?%IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?Q@QBBD7E7GGIJLLNOQvQvSSU\V\XXZB[B]]_(`(bbdeggiikglgnnpMqMssu3v3xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g7 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)F*F44>.?.IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::**  "#%x%x'')d*d,,.P/P113<4<668(9(;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66tt b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2W2W2W2W2-W2-W2X2pX2s[X2s[X2_X2_X2X2`X2oX2oX2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2oO\2T oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  SS)9*944>?IISS]n^nhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..vv\ \""$B%B'')(*(,,./11335g6g88:M;M==?/@/BBDEGGIIKdLdNNPFQFSSU(V(XXZ [ ]{]{__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX23X2`3X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g9 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)>*>44>"?"IIST^x^xhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**r r""$Z%Z'')B*B,,.*/*11346688:n;n==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2aW2aW2W2W2W2W23W23W2X2pX2s[X2s[X2`X2`X2X2`X2cX2cX2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1,L  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  VV)@*@44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99)) u u""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2юW2юW2iW2iW2W2W2W2W2;X2p;X2[X2[X2ubX2ubX2X2`X2{X2{X2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>:?:IIS&T&^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffXX J J  <<..    "#%}%}'')o*o,,.a/a113S4S668E9E;;=7>7@@B)C)EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m o}o}qqsctcvvxIyI{{}/~/kk݌ݎOO b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2aW2aW2W2W2W2W23W23W2X2pX2m[X2m[X2`X2`X2X2`X2{X2{X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1\} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1\} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1\ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>$?$IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::nnߎߐPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2+GW2+GW2ɇW2 ɇW2OW2OW2W2W2W2W2!W2!W2X2pX2m[X2m[X2_X2_X2X2`X2cX2cX2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations11Q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2#Y2 #Y2KZ2p KZ2)[2t )[2P\2T P\2 ]2 ]2'3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yI b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yI b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yI b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)B*B44>0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$j%j'')Z*Z,,.J/J113:4:668*9*;;=>@@B C EEGGIpJpLLN\O\QQSHTHVVX4Y4[[] ^ ``b c eegginjnllndodqqsPtPvvx!?!IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2UX2UX2X2X27NY2L 7NY2 Y2  Y23Z2p 3Z2[2t [2P\2T P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yK b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1GC  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>$?$IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hbbd0e0ggijllnoq|q|ssudvdxxzL{L}}44xx`` b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2%GW2%GW2ÇW2 ÇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2_X2`_X2X2X2 X2 X21NY2L 1NY2Y2 Y2?Z2p ?Z2[2t [2MP\2T MP\2C ]2C ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1X} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1X} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1X } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  YY)G*G44>1?1IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUllnIoIqqs1t1vvxy{{}~uu]]чщEE-- b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2ˎW2ˎW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2obX2obX2_X2`_X2X2X2 X2 X21NY2L 1NY2Y2 Y2?Z2p ?Z2[2t [2SP\2T SP\2I ]2I ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  LL)D*D44>0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln?o?qqs)t)vvxy{{}}rr\\чщFF00 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2a[X2a[X2_X2_X2_X2`_X2X2X2 X2 X21NY2L 1NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Y} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Y} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dx b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Y } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)?*?44>%?%IIS T ^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpaqassuEvExxz){)}} ccՉՋGG++ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2AW2 AW2ˎW2ˎW2uW2uW2W2W2W2W2GX2pGX2[X2[X2obX2obX2_X2`_X2X2X2 X2 X21NY2L 1NY2Y2 Y2-Z2p -Z2 [2t  [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  KK)=*=44>%?%IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##qq] ]""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpeqessuKvKxxz1{1}}ppVVɎɐ<< b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2'W2'W2W2|W2W2HW2W2W2_W2P_W2W2XW2W2W2X2PX2X2X2X2X21[X2$ 1[X2`X2, `X2cX2 cX2UmX2$ UmX2pX2 pX2͝X2| ͝X2X2XX2 X2 X2OX2OX2 Y2 Y22?2IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : : wwee{{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKllnAoAqqs+t+vvxy{{}}tt^^ӇӉHH22# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2YW2dYW2W2lW2W2W2ǛW2ǛW2W2@W2%W2%W2W2W2cX2@cX2X2 X2YX2YX2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2/X2 /X2X2H X2X2hX2Y2DY2= 4 t  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d%Yn47 f?qJ%[  6   l G "  ;/o  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  PP)*44>'?'IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$[%['')E*E,,.///1|1|335f6f88:P;P==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;##mm  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2'W2'W2W2|W2W2HW2W2W2YW2PYW2W2XW2W2W2X2PX2X2X2X2X2+[X2$ +[X2`X2, `X2cX2 cX2UmX2$ UmX2pX2 pX2͝X2| ͝X2ץX2XץX27X27X2yX2yX2Y2Y2'?'IIS T ^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7 pp\\qq] ]""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnpgqgssuMvMxxz3{3}}rrXXˎː>># b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2YW2dYW2W2lW2W2W2ǛW2ǛW2W2@W2+W2+W2W2W2cX2@cX2X2 X2_X2_X2[X2[X2bX2 bX25lX2 5lX2koX2d koX2X2 X2GX2 GX2X2H X2X2hX2Y2DY2O ~ 0\r\ bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)*44>$?$IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$I%I'')3*3,,./1i1i335S6S88:=;===?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggijlulunnpYqYssu=v=xxz!{!}}ww]]ЉЋCC))  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2%GW2%GW2ÇW2dÇW2 W2l W2W2W2͛W2͛W2W2@W2W2W2W2W2iX2@iX2X2 X2X2X2a[X2a[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2X2 X2eX2H eX2X2hX2 Y2D Y2e>Y2e>Y2/_Y2/_Y2-jY2 -jY2Y2Y2Z2pZ2[2t[2[2T[2#P\2#P\2!]2$!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 $dE  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 $dE  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 $dE   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1!aaF  T  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d fnK;0z_D)w^ E ,   z 1  akR bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)>*>44>?IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22""  ""$p%p'')\*\,,.H/H113444668 9 ;;= > @@BBDhEhGGI(J(LLNOQQSSUVXY[|[|]]_d`dbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2MW2dMW2 W2l W2W2W2͛W2͛W2W2@W2W2W2W2W2iX2@iX2X2 X2SX2SX2[X2[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X25X2 5X2X2H X2X2hX2iY2DiY2%    `  0T bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>/?/IIST^^hhrusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnMoMqqs1t1vvxy{_{_}}CC''33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2 W2l W2W2W2͛W2͛W2W2@W2W2W2W2W2iX2@iX2X2 X2X2X2a[X2a[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2X2 X2AX2H AX2X2hX2Y2DY2C>HIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""j j""$R%R''):*:,,."/"113 4 6~6~88:f;f==?L@LBBD0E0GGIIK]L]NNPAQASSUNVNXXZ2[2]]_`bbdegtgtiikXlXnnp>   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2 W2l W2W2W2͛W2͛W2W2@W2%W2%W2W2W2iX2@iX2X2 X2YX2YX2[X2[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2MX2 MX2X2H X2X2hX2'Y2D'Y2;Y2;Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2 [2t [2[2T[2O\2O\2!]2$!]2S4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E OO    b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dE OYLcd(/m2w<   F   P   Z ^  $d bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpgqgssuKvKxxz{}x}x\\΄ΆiiۉۋMM11  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2g[X2g[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2_X20 _X2X2PX2gX2,gX29[Y29[Y2Y2$Y2Z2Z2#[2#[2kP\2kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 3S b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^)t   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  TT)F*F44>8?8IIS$T$^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbTT F F  88**  "#%y%y'')k*k,,.]/]113O4O668A9A;;=3>3@@B%C%EEGHJJLMO~O~QQSlTlVVXZYZ[[]H^H``b4c4eeghjjlmo{o{qqsatavvxGyG{{}-~-kk݌ݎOO b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2W2TW2W2W2ߛW2ߛW2{W2({W2W2W2W2W2{X2({X2X2X2MX2MX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2gX2,gX2][Y2][Y2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d1~ '  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2%GW2%GW2ÇW2LÇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2g[X2g[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X25X2 5X2X20 X2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 xGgl b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1? U N  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbdWeWggi=j=lln!o!qqstvtvtxxzV{V}}88nnߎߐPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2SX2SX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2YX2 YX2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ke2t  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2GX20 GX2X2PX2eX2,eX2[Y2[Y2Y2$Y2Z2Z2)[2)[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 YG  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 YG  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 YG   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ; b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Qc  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)B*B44>2?2IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$j%j'')Z*Z,,.J/J113:4:668*9*;;=>@@B C EEGGIrJrLLN^O^QQSJTJVVX6Y6[[]"^"``bceeggipjplln\o\qqsHtHvvx4y4{{} ~  nnZZБГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW27W2T7W2W2W2W2W2uW2(uW2W2W2 W2 W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2}X2 }X2X20 X2X2PX2aX2,aX2{[Y2{[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Z. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Z. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Z.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 "B b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1],{ >  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2MX2 MX2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 XF bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 XF bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 XF  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,u>! a  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>&?&IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqvqvssu^v^xxzF{F}}..rrZZ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW27W2T7W2W2W2W2W2uW2(uW2W2W2 W2 W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2CX20 CX2X2PX2X2,X2 \Y2 \Y2Y2$Y2Z2Z2[2[2iO\2iO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 @`3 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Jh7  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2[X2,[X2o[Y2o[Y2Y2$Y2Z2Z2#[2#[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I]@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I]@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I]@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Nj 0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2 X2  X2mX20 mX2X2PX2aX2,aX2[Y2[Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Z" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Z" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Z"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 'G5U b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1V'x U  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  TT)>*>44>)?)IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''s s""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2UX20 UX2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J\> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J\> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J\>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $D b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18T"~  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2ѥX2 ѥX21X20 1X2sX2PsX2Y2,Y2#\Y2#\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Jj b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ik<  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2W2TW2W2W2ٛW2ٛW2 W2( W2W2W2W2W2uX2(uX2X2X2X2X2y[X2y[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2٦X2 ٦X29X20 9X2{X2P{X2kX2,kX2#\Y2#\Y2Y2$Y2Z2Z2#[2#[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NE  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NE  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NE   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 >^! b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 L bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1[4   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLN3?3IIS!T!^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGI[J[LLNIOIQQS7T7VVX%Y%[[]^``bcexexggifjfllnOoOqqs3t3vvxy{{}}mm߂߄QQÇÉ55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2s[X2s[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X27X2P7X2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  ME  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  ME  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  ME   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1f?q;{ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2W2TW2W2W2ٛW2ٛW2iW2(iW2W2W2W2W2uX2(uX2X2X2;X2;X2[X2[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2GX2 GX2X20 X2X2PX2X2,X2}\Y2}\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Lll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\3 a 7 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2W2TW2W2W2ߛW2ߛW2 W2( W2W2W2W2W2{X2({X2X2X2X2X2y[X2y[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2+X2 +X2X20 X2X2PX2aX2,aX2\Y2\Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @F  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @F  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @F   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .NNn b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1dH,P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ߛW2ߛW2]W2(]W2W2W2W2W2{X2({X2X2X2/X2/X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2wX20 wX2X2PX2aX2,aX2\Y2\Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;;[ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1bD&P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2IX20 IX2X2PX2Y2,Y2y]Y2y]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @F bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @F bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @F  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  / b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tX<4 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2W2TW2W2W2ߛW2ߛW2cW2(cW2W2W2W2W2{X2({X2X2X25X25X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2Y2,Y27]Y27]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 >^^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jL. } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>)?)IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpYqYssu=v=xxz!{!}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2OX20 OX2X2PX2qX2,qX2]Y2]Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 CE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 CE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 CE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f / b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1o^M@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2%W2T%W2W2W2W2W2WW2(WW2W2W2W2W2X2(X2X2X2)X2)X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2GX2 GX2X20 X2X2PX2gX2,gX2I]Y2I]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mZG } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?Q@QBBD7E7GGIJLLNOQvQvSSU\V\XXZB[B]]_(`(bbdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2[X2 [X2X20 X2X2PX2Y2,Y29^Y29^Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 BE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 BE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 BE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[[{ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<+  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2%W2T%W2W2W2W2W2]W2(]W2W2W2W2W2X2(X2X2X2/X2/X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2qX20 qX2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2 [2 [2{O\2{O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 H  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 H  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S'  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 H   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 /4T b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14!{m } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  QQ)7*744>?IISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?/@/BBDEGGIIKdLdNNPFQFSSU(V(XXZ [ ]{]{__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2[W2$[W29W29W2W2W2-W2$-W2 X2 X2[X2[X2_X2$_X2X2X2{X2{X2'X2 'X2ONY2| ONY2;Y2 ;Y2Z2 Z2M[2 M[2qP\2 qP\2A"]2A"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdnenggiXjXllnBoBqqs,t,vvxy{{}~uu__ԇԉII33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2QW2QW2W2W2W2W2#X2p#X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  aa)K*K44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44!!   ""$j%j'')T*T,,.>/>113(4(6689;;==?q@qBBD[E[GGIDJDLLN,O,QQSTVVXXZp[p]]_X`Xbbd@e@ggi(j(llnoqqssulvlxxzT{T}}<<$$ ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2[W2$[W2?W2?W2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2ONY2| ONY2#Y2 #Y2Z2 Z25[2 5[2P\2 P\2!]2!]2}4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)L*L44>4?4IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI8J8LLNOQQSTVwVwXXZ][]]]_C`Cbbd)e)ggijllnnphqhssuNvNxxz4{4}}ssYY̎̐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ey b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  __)I*I44>1?1IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI1J1LLNOQQSSUpVpXXZV[V]]_<`v>xxz"{"}}xx\\ωϋBB(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2[W2[W2!W2!W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2cX2cX2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  gg)S*S44>;?;IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG77##  "#%%'')q*q,,.]/]113I4I668595;;=>@@BCE{E{GGIcJcLLNKOKQQS3T3VVXY[[]^`w`wbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2WW2WW2W2W2W2W2)X2p)X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>:?:IIS(T(^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGIbJbLLNPOPQQS>T>VVX,Y,[[]^``bceeggimjmllnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2[W2[W2!W2!W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1B} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1B} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1B } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  aa)I*I44>-?-IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55!!  } }""$e%e'')M*M,,.5/511346689;y;y==?]@]BBDAEAGGI%J%LLN O Q{Q{SSU_V_XXZC[C]]_'`'bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N" b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ZZ)D*D44>.?.IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppZZ D D  ..wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBDG?GIIS3T3^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssee W W  II;;--  "#%%'(*|*|,,.n/n113`4`668R9R;;=D>D@@B6C6EEG(H(JJLMOOQRT{T{VVXiYi[[]W^W``bCcCeeg-h-jjlmooqqsptpvvxVyV{{}<~<""xx\\ΑΓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W) b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  dd)P*P44>:?:IIS"T"^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@44((  "#%%'(*~*~,,.j/j113V4V668B9B;;=.>.@@BCEEGHJzJzLLNbObQQSJTJVVX2Y2[[]^``bcevevggi^j^llnFoFqqs.t.vvxy{{}}rrZZ·ΉBB** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS!T!^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbddedggiJjJlln.o.qqstvvxxzc{c}}EE''yy[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)H*H44>.?.IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00    "#%t%t'')\*\,,.D/D113,4,6689;;==?p@pBBDXEXGGI>J>LLN"O"QQSTVxVxXXZ\[\]]_@`@bbd$e$ggijlzlznnp^q^ssuBvBxxz&{&}} ||``҉ҋDD(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2gW2gW2W2W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2X2X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Z} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Z} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Z } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  __)O*O44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??//  ""$w%w'')g*g,,.W/W113G4G668797;;='>'@@BCEEGHJ}J}LLNiOiQQSUTUVVXAYA[[]-^-``bceeghj{j{llngogqqsStSvvx?y?{{}+~+yyeeۑۓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2ݎW2ݎW2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`2 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  TT)>*>44>$?$IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiUU A A  --yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2mW2mW2W2W2W2W2?W2?W2X2pX2[X2[X2`X2`X2٣X2`٣X29X29X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)E*E44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$W%W'')C*C,,.///11346689;};}==?i@iBBDUEUGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77{{cc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2ݎW2ݎW2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX29X2`9X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  TT):*:44>?IISS]o^ohhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //ww] ]""$C%C''))*),,./11335h6h88:N;N==?0@0BBDEGGIIKeLeNNPGQGSSU)V)XXZ [ ]|]|__a^b^ddf@g@iik"l"nnpqsusuuuwWxWzz|9}9kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2oX2oX2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbdbebggiLjLlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2'X2`'X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i; b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2 W2 W2W2W2-W2-W2X2pX2y[X2y[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)@*@44>(?(IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$S%S'')=*=,,.'/'11346688:p;p==?Z@ZBBDDEDGGI,J,LLNOQQSSUkVkXXZQ[Q]]_7`7bbdeggijlvlvnnp\q\ssuBvBxxz({(}}ggډڋMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2?X2`?X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2aW2aW2 W2 W2W2W23W23W2X2pX2y[X2y[X2`X2`X2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>/?/IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbdSeSggi;j;lln#o#qqs t vvxxzg{g}}OOÂÄ77uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2-X2`-X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rD b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Uu  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2aW2aW2W2W2W2W23W23W2X2pX2s[X2s[X2`X2`X2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1X} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1X} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1X } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  UU)=*=44>!?!IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2юW2юW2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2ubX2ubX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rF b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2OW2OW2W2W2W2W2!W2!W2X2pX2s[X2s[X2_X2_X2X2`X2WX2WX2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1%E  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>;?;IIS'T'^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2?Z2p ?Z2[2t [2MP\2T MP\2C ]2C ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  XX)D*D44>.?.IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44((  "#%%'')r*r,,.^/^113J4J668696;;=">"@@BCEEGGInJnLLNVOVQQS>T>VVX&Y&[[]^``bbdjejggiRjRllnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2+GW2+GW2ɇW2 ɇW2OW2OW2W2W2W2W2!W2!W2X2pX2m[X2m[X2_X2_X25X2`5X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1`  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX27X2`7X2X2X2X2X27NY2L 7NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>"?"IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI2J2LLNOQQSSUlVlXXZP[P]]_4`4bbdeggiiknlnnnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2+GW2+GW2ɇW2 ɇW2UW2UW2W2W2W2W2'W2'W2X2pX2g[X2g[X2_X2_X2X2`X2KX2KX2 X2 X21NY2L 1NY2#Y2 #Y2KZ2p KZ2)[2t )[2P\2T P\2 ]2 ]2'3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1T} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1T} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1T } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1B>  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)E*E44>3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqllnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2GW2 GW2ˎW2ˎW2uW2uW2W2W2W2W2GX2pGX2[X2[X2obX2obX2X2`X2X2X2 X2 X21NY2T 1NY2Y2 Y2'Z2x 'Z2[2| [2O\2\ O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V" b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1%  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10]]  RR)<*<44>"?"IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1]]II 5 5  !!  wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbdeggiikplpnnpVqVssu'?'IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2GW2 GW2ˎW2ˎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2obX2obX2)X2`)X2X2X2 X2 X21NY2T 1NY2Y2 Y2Z2x Z2[2| [2oO\2\ oO\2e]2e]2 2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1{  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  PP)6*644>?IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2-W2-W2W2|W2W2HW2W2W2_W2P_W2W2XW2W2W2X2PX2X2X2X2X21[X2$ 1[X2`X2, `X2cX2 cX2[mX2$ [mX2pX2 pX2ӝX2| ӝX2X2XX2X2X2+X2+X2 Y2 Y2>Y2\>Y2\Y2\Y2gY2gY2Y28Y2Z2Z2)[2)[2[2[2MP\2lMP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P{  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1-m  F   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dGUc*q8Hh7V    d +  r  @  N/o bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa UU)K*K44>7?7IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I %%  " # %%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnOoOqqs7t7vvxy{{}~{{ccׇ׉KK33# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2YW2dYW2 W2l W2W2W2͛W2͛W2W2@W2+W2+W2W2W2iX2@iX2X2 X2_X2_X2[X2[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2AX2 AX2X2H X2X2hX2C Y2DC Y2=Y2=Y2a]Y2a]Y2_hY2 _hY2Y2Y2Z2pZ2/[2t/[2[2T[2wP\2wP\2G"]2$G"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 | b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1N"b b  2 b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dNj8T"p\@N`0   P  p > ~  \, bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV) * 44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsqtqvvxWyW{{}=~=## {{__ёѓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2-W2-W2W2|W2W2HW2W2W2_W2P_W2W2XW2W2W2X2PX2X2X2X2X21[X2$ 1[X2`X2, `X2cX2 cX2[mX2$ [mX2pX2 pX2ӝX2| ӝX2X2XX2yX2yX2X2X2' Y2' Y2>Y2\>Y27]Y27]Y25hY25hY2Y28Y2Z2Z2[2[2[2[2O\2lO\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ox b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ox b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ox  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0PPp b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1N: I   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dNj8T"p<T  l U  8 $   s P B  ^  8x bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa QQ)A*A44>)?)IIS T ^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A ~~nn y y""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpeqessuIvIxxz-{-}}ggىًKK//# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2_W2d_W2 W2l W2W2W2͛W2͛W2W2@W21W21W2W2W2iX2@iX2X2 X2eX2eX2\X2\X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2MX2 MX2X2H X2X2hX2Y2DY20?0IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$_%_'')M*M,,.;/;11335u6u88:c;c=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ~~``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2%GW2%GW2ÇW2dÇW2 W2l W2W2W2͛W2͛W2W2@W2W2W2W2W2iX2@iX2X2 X2X2X2g[X2g[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2X2 X2qX2H qX2X2hX2Y2DY2g>HIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^JJ 6 6  ""nnX X""$B%B''),*,,,./11346u6u88:_;_==?G@GBBD-E-GGIIK\L\NNPBQBSSURVRXXZ8[8]]_`bbdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2SW2dSW2 W2l W2W2W2͛W2͛W2W2@W2%W2%W2W2W2iX2@iX2X2 X2YX2YX2[X2[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2GX2 GX2X2H X2X2hX2Y2DY2= ~  L bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)D*D44>4?4IIS T ^^hhrs} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln`o`qqsLtLvvx8y8{{}}pp\\rr^^ԑԓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2%GW2%GW2ÇW2dÇW2 W2l W2W2W2͛W2͛W2W2@W2W2W2W2W2iX2@iX2X2 X2X2X2g[X2g[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2ϤX2 ϤX2/X2H /X2qX2hqX2Y2DY2>HISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$llR R""$8%8'')*,,./1w1w335]6]88:C;C==?'@'BBD E GzGzIIK1L1NNPQSSU V XXZ[]s]s__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2YW2dYW2 W2l W2W2W2͛W2͛W2W2@W2+W2+W2W2W2iX2@iX2X2 X2_X2_X2[X2[X2bX2 bX2;lX2 ;lX2qoX2d qoX2X2 X2YX2 YX2X2H X2X2hX2 Y2D Y2@Y2@Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2[2t[2[2T[2P\2P\2!]2$!]2}4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1GU  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dGUc*q8F T U    b L )   p  > ~ 3 s  L bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>&?&IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqzqzssubvbxxz{}}{{vv^^  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2GX2PGX2uY2,uY2M\Y2M\Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TG bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TG bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TG  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1!a  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1i4J  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)8*844>#?#IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!m m""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI3J3LLNOQQSTVwVwXXZ_[_]]_G`Gbbd/e/ggijllnnpsqsssu[v[xxzC{C}}++ss]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2SX2SX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2SX2 SX2X20 X2X2PX2WY2,WY2E[Y2E[Y2 Y2$ Y2Z2Z2[2[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1G  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1m*g$   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2+GW2+GW2ɇW2LɇW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2AX2 AX2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UG bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UG bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UG  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Rrw b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ` + v b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)6*644> ? IIST^w^whhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikelennpIqIssu-v-xxz{}}ggلنMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2SX2SX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2_X2 _X2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1a  [  L  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2kX2 kX2X20 X2 X2P X2uY2,uY2#\Y2#\Y2 Y2$ Y2Z2Z2[2[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UI  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UI  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UI   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1N`)  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIUJULLNCOCQQS1T1VVXY[[] ^ ``bbdrerggi`j`llnIoIqqs-t-vvxy{{}}ggقلKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW27W2T7W2W2W2W2W2{W2({W2W2W2 W2 W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2yX2 yX2X20 X2X2PX2_Y2,_Y2[Y2[Y2 Y2$ Y2Z2Z2[2[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .N b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(k.qJ  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2m[X2m[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2YX20 YX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TH bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TH bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TH  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 +K b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1r ;  M 1q } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW27W2T7W2W2W2W2W2{W2({W2W2W2 W2 W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X27X2P7X2Y2,Y2G\Y2G\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Vv b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14 w : }  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2OW2TOW2٘W2٘W2W2W2 W2( W2W2W2!W2!W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2mY2,mY2\Y2\Y2Y2$Y2Z2Z2#[2#[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I_@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I_@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I_@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 R  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p@` L } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  YY)E*E44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')s*s,,._/_113K4K668797;;=#>#@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2WX2PWX2oY2,oY2\Y2\Y2#Y2$#Y2Z2Z25[25[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 8X b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Nj P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  ZZ)L*L44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhZZ L L  >>00""  "#%%'')q*q,,.c/c113U4U668G9G;;=9>9@@B+C+EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3qqUUǑǓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J^> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J^> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J^>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 { b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 0 P  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2CX2PCX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 \| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 8  T "  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2W2TW2W2W2ߛW2ߛW2 W2( W2W2W2W2W2{X2({X2X2X2X2X2y[X2y[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2X20 X2X2PX2Y2,Y2[]Y2[]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JG bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JG bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JG  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $DDd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1>~  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1J% [  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbdeg{g{iikalannpGqGssu-v-xxz{}}ll߄߆RRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ߛW2ߛW2iW2(iW2W2W2W2W2{X2({X2X2X2;X2;X2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2;X2 ;X2X20 X2X2PX2Y2,Y2[]Y2[]Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i=   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bbb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations16v bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X1 c  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>8?8IIS$T$^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJxJxLLNdOdQQSPTPVVX?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2W2TW2W2W2ߛW2ߛW2oW2(oW2 W2 W2W2W2{X2({X2X2X2AX2AX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2eX2 eX2X20 X2X2PX2/Y2,/Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<  n G o } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2SX2 SX2X20 X2X2PX2sY2,sY2!^Y2!^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @H bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @H bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @H  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Uuz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 R  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 l R  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2%W2T%W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2AX2 AX2X20 X2X2PX2Y2,Y2'^Y2'^Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ddd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1-m  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10x \  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKlln5o5qqstvvx y {~{~}}hh݂݄RRLJlj<<&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2mX2 mX2X20 X2X2PX2 Y2, Y2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @H bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @H bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @H  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 8   j >~ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2%W2T%W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X2=X2P=X23Y2,3Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 @ $  l @ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>'?'IIS T ^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnp[q[ssuAvAxxz'{'}} ffىًLL22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2OX20 OX2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2/[2/[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @H bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @H bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @H  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  1 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1h  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xiZ K ^ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2)W2L)W2%W2T%W2W2W2W2W2]W2(]W2W2W2W2W2X2(X2X2X2/X2/X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2kX2 kX2X20 X2 X2P X2Y2,Y2^Y2^Y2#Y2$#Y2Z2Z25[25[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S'  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1wf U b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<,,  "#%z%z'')f*f,,.R/R113>4>668*9*;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2SX2 SX2X20 X2X2PX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @H bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @H bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @H  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Oot b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 Y  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1< -  O } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2%W2T%W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2qX20 qX2X2PX2wY2,wY2_Y2_Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ,1Q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D 3 " Q } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)>*>44>$?$IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**r r""$Z%Z'')B*B,,.*/*11346688:n;n==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp>  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2W2[W2$[W23W23W2W2W2-W2$-W2X2X2[X2[X2_X2$_X2'X2'X2X2X2'X2 'X2ONY2| ONY2;Y2 ;Y2Z2 Z2M[2 M[2qP\2 qP\2A"]2A"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:Z  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  nn)Z*Z44>D?DIIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ>>22  "&#&%%'(**,,.t/t113`4`668L9L;;=8>8@@B$C$EEGHJJLLNlOlQQSTTTVVXI?IIIS5T5^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uugg Y Y  KK==//!!  "#%%'(*~*~,,.p/p113b4b668T9T;;=F>F@@B8C8EEG*H*JJLMOOQRT}T}VVXkYk[[]Y^Y``bEcEeeg/h/jjlmooqqsrtrvvxXyX{{}>~>$$zz^^БГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2W2[W2$[W29W29W2W2W2-W2$-W2 X2 X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2ONY2| ONY2#Y2 #Y2Z2 Z25[2 5[2O\2 O\2!]2!]2q4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)P*P44>6?6IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%|%|'')d*d,,.L/L1134446689;;=>@x@xBBD`E`GGIFJFLLN*O*QQSTVVXXZd[d]]_H`Hbbd,e,ggijllnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  cc)Q*Q44>=?=IIS%T%^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uucc Q Q  ??--   ""$n%n'')\*\,,.J/J113848668&9&;;=>@@BCEyEyGGIeJeLLNOOOQQS9T9VVX#Y#[[] ^ ``bbdhehggiNjNlln2o2qqstvvxxzg{g}}II++ }}__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ^^)H*H44>.?.IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##m m""$W%W'')A*A,,.+/+11346688:t;t==?Z@ZBBD@E@GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpVqVssu???IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11!!  "#%y%y'')i*i,,.Y/Y113I4I668999;;=)>)@@BCEEG H JJLLNkOkQQSWTWVVXCYC[[]/^/``bceeghj}j}llnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2WX2WX2'X2'X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2W!]2W!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations10  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  \\)B*B44>$?$IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##  e e""$K%K'')1*1,,./11335p6p88:V;V==?8@8BBDEGGIIKmLmNNPOQOSSU1V1XXZ[]]__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2QX2QX2'X2'X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 )  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])I*I44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$[%['')G*G,,.3/31134668 9 ;;==?m@mBBDYEYGGICJCLLN+O+QQSTVVXXZo[o]]_W`Wbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2[W2[W2!W2!W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2cX2cX2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1>  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  __)I*I44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22  ~ ~""$h%h'')R*R,,.e>ggi&j&llnoqqssujvjxxzR{R}}::"" jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>5?5IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI]J]LLNGOGQQS1T1VVXY[[]^`z`zbbddedggiNjNlln8o8qqs"t"vvx y {{}}kkUUʇʉ??)) b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>-?-IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?[@[BBDEEEGGI-J-LLNOQQSSUlVlXXZR[R]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXˉˋ>>$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2KX2KX2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1(  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ZZ)D*D44>,?,IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  l l""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI0J0LLNOQQSSUoVoXXZU[U]]_;`;bbd!e!ggijlzlznnp`q`ssuFvFxxz,{,}}kkމދQQĎĐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2gW2gW2W2W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1V} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1V} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1V } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>8?8IIS&T&^^h i s|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI`J`LLNNONQQS1?1IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==--  "#%{%{'')g*g,,.S/S113?4?668+9+;;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99ww b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2gW2gW2W2W2W2W29W29W2X2pX2[X2[X2 `X2 `X2٣X2`٣X29X29X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  VV)@*@44>*?*IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llVV @ @  **ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLN O QQSSUkVkXXZU[U]]_?`?bbd)e)ggijllnnplqlssuPvPxxz4{4}}nnRRĎĐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)A*A44>%?%IIS T ^{^{hhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==-- u u""$]%]'')E*E,,.-/-11346688:q;q==?U@UBBD9E9GGIJLLNOQsQsSSUWVWXXZ;[;]]_`bbdeguguiikYlYnnp=q=ssu!v!xxz{}w}w[[̈́͆== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>22&&  "#%%'(*|*|,,.h/h113T4T668@9@;;=,>,@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX29X2`9X2X2X2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k?  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k?  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k?  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1[{  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  [[)M*M44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii[[ M M  ??11##  "#%%'')r*r,,.d/d113V4V668H9H;;=:>:@@B,C,EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)D*D44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$p%p'')X*X,,.@/@113(4(6689;;==?l@lBBDTETGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>"?"IIST^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2׎W2׎W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2{bX2{bX2kX2`kX2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tH b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2[W2[W2 W2 W2W2W2-W2-W2X2pX2y[X2y[X2_X2_X2 X2` X2iX2iX2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2oO\2T oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1T} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1T} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1T } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations10P  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]k^khhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2UW2UW2W2W2W2W2'W2'W2X2pX2s[X2s[X2_X2_X2X2`X2QX2QX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>&?&IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2}X2`}X2X2X2X2X27NY2L 7NY2Y2 Y2?Z2p ?Z2[2t [2MP\2T MP\2C ]2C ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>)?)IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIQJQLLN;O;QQS%T%VVXY[[]]_n`nbbdXeXggiBjBlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W21GW21GW2χW2 χW2OW2OW2W2W2W2W2!W2!W2X2pX2s[X2s[X2_X2_X2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2)X2`)X2X2X2X2X27NY2L 7NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1{w  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844> ? IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2GW2 GW2ˎW2ˎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2obX2obX2ǣX2`ǣX2'X2'X2X2X21NY2T 1NY2Y2 Y29Z2x 9Z2[2| [2/P\2\ /P\2% ]2% ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1c  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)G*G44>/?/IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11!!   ""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2+GW2+GW2ɇW2 ɇW2OW2OW2W2W2W2W2!W2!W2X2pX2m[X2m[X2_X2_X2X2`X2X2X2X2X21NY2L 1NY2Y2 Y2-Z2p -Z2 [2t  [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1H| b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)8*844>"?"IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnppqpssuTvTxxz8{8}}rrVVȎȐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2MW2 MW2ˎW2ˎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2obX2obX2X2`X2wX2wX2X2X21NY2T 1NY2Y2 Y2!Z2x !Z2[2| [2O\2\ O\2]2]2U2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1i  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)=*=44>!?!IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!  i i""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W2-W2-W2W2|W2W2HW2ǰW2ǰW2eW2PeW2W2XW2W2W2X2PX2X2X2X2X27[X2$ 7[X2`X2, `X2cX2 cX2[mX2$ [mX2pX2 pX2ӝX2| ӝX2X2XX2X2X2QX2QX29Y29Y2;Y2\;Y2\Y2\Y2gY2gY2#Y28#Y2Z2Z25[25[2[2[2P\2lP\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q/ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1OOa  *  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dIN [L$m6HZ#l  5 ~ G  ?   b  +t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ OO)I*I44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ? {{kkw w""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllnoooqqs[t[vvxGyG{{}3~3 mm# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2_W2d_W2W2lW2W2W2ӛW2ӛW2W2@W21W21W2W2W2oX2@oX2X2 X2eX2eX2\X2\X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2SX2 SX2X2H X2X2hX2?Y2D?Y2;Y2;Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2#[2t#[2[2T[2;P\2;P\2 "]2$ "]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1QQc  ,  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dIJ]<&o8J\%n  7  I    d  -v bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)*44>0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/911335r6r88:`;`=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnXoXqqs>""  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W2-W2-W2W2|W2W2HW2W2W2_W2P_W2W2XW2W2W2X2PX2X2X2X2X21[X2$ 1[X2`X2, `X2cX2 cX2[mX2$ [mX2pX2 pX2ӝX2| ӝX2X2XX2X2X27X27X2]Y2]Y2+?+IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7 llXXqq] ]""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoqqssuovoxxzW{W}}??''kk# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2eW2deW2W2lW2W2W2ӛW2ӛW2W2@W27W27W2W2W2oX2@oX2X2 X2kX2kX2 \X2 \X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2AX2 AX2X2H X2X2hX2QY2DQY2 $?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$?%?''))*),,./1[1[335E6E88:/;/==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2+GW2+GW2ɇW2dɇW2W2lW2W2W2ӛW2ӛW2W2@W2W2W2W2W2oX2@oX2X2 X2X2X2m[X2m[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2)X2 )X2X2H X2X2hX2Y2DY2=Y2=Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2#[2t#[2[2T[2GP\2GP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p>^^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1I3s x  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dIK[E$m6HZ#l  5 ~ G     b  +t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  MM);*;44>)?)IrIrST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGIQJQLLN;O;QQS%T%VnVnXXZX[X]]_B`BbbdXeXggiBjBlln6o6qqs t vvx y {{}}iiނބSSȇȉ==''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2YW2dYW2W2lW2W2W2ӛW2ӛW2W2@W2+W2+W2W2W2oX2@oX2X2 X2_X2_X2[X2[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2ۤX2 ۤX2;X2H ;X2}X2h}X2Y2DY22?2IIST^^hhrxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln@o@qqs(t(vvxy{X{X}}@@((ȇȉ<<$$  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2+GW2+GW2ɇW2dɇW2W2lW2W2W2ӛW2ӛW2W2@W2W2W2W2W2oX2@oX2X2 X2X2X2m[X2m[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2/X2 /X2X2H X2X2hX2SY2DSY2=Y2=Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2 [2t [2[2T[2O\2O\2!]2$!]2S4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ddi b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1IZZ  ?  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dI[$m?6|9HZ#l  5 ~ G  \   b  +t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  LL)6*644> ? IfIfST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSU6V6XXZ[]]_`bbdeggiiknlnnnp\q\ssuBvBxxz({(}}ggډڋMM33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2_W2d_W2W2lW2W2W2ӛW2ӛW2W2@W2+W2+W2W2W2oX2@oX2X2 X2_X2_X2[X2[X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2;X2 ;X2X2H X2X2hX2 Y2D Y2>Y2>Y2\Y2\Y2gY2 gY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2G4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1r  7   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dI[$xm36o*HZ#l  5 ~ G  ; {   b  +t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>$?$IIST^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpXqXssu>""  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2s[X2s[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2)X2P)X2UX2,UX2IZY2IZY2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TI bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TI bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TI  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<x4   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)?*?44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33##  ""$q%q'')]*],,.I/I113545668!9!;;= > @@BBDiEiGGIQJQLLN9O9QQS!T!VVX Y [}[}]]_e`ebbdMeMggi5j5llnoqqstvyvyxxza{a}}II11qq b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2W2TW2W2W2ߛW2ߛW2W2(W2%W2%W2W2W2{X2({X2X2X2YX2YX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2wX2 wX2X20 X2X2PX2UX2,UX2OZY2OZY2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ff b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1C bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@}:   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W21GW21GW2χW2LχW2W2TW2W2W2ٛW2ٛW2W2(W2W2W2W2W2uX2(uX2X2X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2L }oX2X2 X2X2 X2X20 X2GX2PGX2X2,X2ZY2ZY2Y2$Y2Z2Z2[2[2O\2O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UI bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UI bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UI  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1| bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p,h$L  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)7*744>?IIST^s^shhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$S%S'');*;,,.#/#113 4 6688:g;g==?M@MBBD1E1GGIJLLNNPkQkSSUOVOXXZ3[3]]_`bbddfmgmiikQlQnnp5q5ssuvxxzz|o}oSSńņ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2W2TW2W2W2ߛW2ߛW2W2(W2%W2%W2W2W2{X2({X2X2X2YX2YX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2OX20 OX2X2PX2X2,X2ZY2ZY2Y2$Y2Z2Z2[2[2cO\2cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :!A b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1w4q.P  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  LL)2*244>?IISS]i^ihhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ooU U""$;%;'')!*!,,./1z1z335`6`88:F;F==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3gg׋׍GG b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2+W2T+W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2X20 X2/X2P/X2X2,X2ZY2ZY2Y2$Y2Z2Z2)[2)[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 QK  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 QK  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 QK   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1*j bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1GU  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>9?9IIS%T%^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMOOQQSmTmVVX[Y[[[]I^I``b5c5eeghjjl m o|o|qqsbtbvvxHyH{{}.~.llތގPP‘“ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW27W2T7W2W2W2W2W2{W2({W2W2W2 W2 W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2AX2 AX2X20 X2X2PX2UX2,UX2ZY2ZY2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Lll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1E O@  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2s[X2s[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2/X2 /X2X20 X2X2PX2 Y2, Y2[Y2[Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 PJ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 PJ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 PJ  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 8X]} b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1c*q G  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln"o"qqstvuvuxxzW{W}}99ooQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2=W2T=W2ǘW2ǘW2W2W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2OX20 OX2X2PX2X2,X2E[Y2E[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1['  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 <9 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Yc   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>#?#IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikhlhnnpLqLssu0v0xxz{}}jj܄܆PPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2OW2TOW2٘W2٘W2W2W2 W2( W2W2W2!W2!W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2SX20 SX2X2PX2[X2,[X2[Y2[Y2#Y2$#Y2Z2Z25[25[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ia@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ia@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ia@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 z!A b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R$v0 p  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)C*C44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')[*[,,.K/K113;4;668+9+;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW21W2T1W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2+X2P+X2[X2,[X2o[Y2o[Y2Y2$Y2Z2Z2)[2)[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Rr b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P p 0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2OW2TOW2٘W2٘W2W2W2 W2( W2W2W2!W2!W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2X2,X2A\Y2A\Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J`> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J`> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J`>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Hl>  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$M%M'')9*9,,.%/%11346688:s;s==?_@_BBDKEKGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW21W2T1W2W2W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2wX20 wX2X2PX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 >>^ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@`0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2eX2 eX2X20 X2X2PX2eX2,eX2e\Y2e\Y2Y2$Y2Z2Z2)[2)[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 FI bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 FI bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 FI  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1H bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]:  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln0o0qqstvvxy{y{y}}cc؂؄MM‡‰77!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2%W2T%W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2mX2 mX2X20 X2X2PX2aX2,aX2#\Y2#\Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kA   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1[6  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJvJvLLN^O^QQSFTFVVX.Y.[[]^``bbdrerggiZjZllnBoBqqs*t*vvxy{{}}nnVVʇʉ>>&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2kX2 kX2X20 X2 X2P X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 EI bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 EI bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 EI  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tQ. } } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}cc։֋II// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2%W2T%W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2}X20 }X2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  @@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lG"};{ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)@*@44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$l%l'')T*T,,.>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2yX20 yX2X2PX2YX2,YX2\Y2\Y2Y2$Y2Z2Z2)[2)[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0P7W b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations15 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1aB# ` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2qX2 qX2X20 X2X2PX2aX2,aX2\Y2\Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`6  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fL2P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2X20 X2CX2PCX2Y2,Y2I]Y2I]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Vv b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1eF' } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>!?!IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2+W2T+W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2#X2 #X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2 [2 [2{O\2{O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b:  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 "BGg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~dJ6 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]m^mhhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?.@.BBDEGGIIKcLcNNPEQESSU'V'XXZ [ ]z]z__a\b\ddf>g>iik l nnpqssssuuwUxUzz|7}7kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2UX2 UX2X20 X2X2PX2gX2,gX2[]Y2[]Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ppu b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lXD } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2/W2L/W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2wX20 wX2X2PX2gX2,gX2]Y2]Y2Y2$Y2Z2Z2/[2/[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 49Y b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1qbS@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  WW)A*A44>,?,IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::** v v""$`%`'')J*J,,.4/411346689;};}==?g@gBBDQEQGGI1?1IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W2+W2T+W2W2W2W2W2cW2(cW2W2W2W2W2X2(X2X2X25X25X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2MX2 MX2X20 X2X2PX2Y2,Y2?^Y2?^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D5& } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>'?'IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnTTljNj::  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2W2[W2$[W23W23W2W2W2-W2$-W2X2X2[X2[X2_X2$_X2X2X2X2X2'X2 'X2ONY2| ONY2GY2 GY2Z2 Z2Y[2 Y[2P\2 P\2}"]2}"]2%5]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)X*X44>F?FIIS2T2^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%%'')p*p,,.`/`113P4P668@9@;;=0>0@@B C EEGHJJLLNrOrQQS^T^VVXJYJ[[]6^6``b"c"eeghjjllnpopqqs\t\vvxHyH{{}4~4  nn b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2 X2` X2iX2iX2'X2'X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I} b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1>  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>=?=IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIeJeLLNSOSQQSATAVVX/Y/[[]^``b c eeggipjpllnYoYqqs=t=vvx!y!{{}~ww[[͇͉??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2W2[W2$[W23W23W2W2W2-W2$-W2X2X2[X2[X2_X2$_X23X23X2X2X2'X2 'X2ONY2| ONY2/Y2 /Y2Z2 Z2A[2 A[25P\2 5P\2"]2"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$`%`'')L*L,,.8/8113$4$6689;;==?r@rBBD^E^GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2]W2]W2W2W2W2W2/X2p/X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}! b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ]])G*G44>1?1IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss]] G G  11zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUrVrXXZ\[\]]_F`Fbbd0e0ggijllnoqsqsssuWvWxxz;{;}}uuYYˎː== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2!X2`!X2X2X2'X2'X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:Z  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdkekggiUjUlln?o?qqs)t)vvxy{{}}rr\\чщFF00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  gg)S*S44>=?=IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC77++  "#%%' ( **,,.m/m113Y4Y668E9E;;=1>1@@BCEEG H J}J}LLNeOeQQSMTMVVX5Y5[[]^``bceyeyggiajallnIoIqqs1t1vvxy{{}~uu]]чщEE-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)I*I44>1?1IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI5J5LLNOQQSTVtVtXXZZ[Z]]_@`@bbd&e&ggi j llnnpeqessuKvKxxz1{1}}ppVVɎɐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2WX2WX2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)K*K44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33##  "#%w%w'')_*_,,.G/G113/4/6689;;==?s@sBBD[E[GGIAJALLN%O%QQS T V{V{XXZ_[_]]_C`Cbbd'e'ggi j l}l}nnpaqassuEvExxz){)}} ccՉՋGG++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2cX2cX2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)P*P44>8?8IIS T ^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD44   " # %%'')n*n,,.Z/Z113F4F668292;;=>@@BCExExGGI`J`LLNHOHQQS0T0VVXY[[]^`t`tbbd\e\ggiDjDlln,o,qqstvvxxzp{p}}XX̂̄@@%% ~~ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  WW)A*A44>'?'IIS T ^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llXX D D  00||f f""$P%P''):*:,,.$/$11346688:m;m==?S@SBBD9E9GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiikilinnpOqOssu5v5xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ^^)F*F44>*?*IIST^^hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22 z z""$b%b'')J*J,,.2/211346689;v;v==?Z@ZBBD>E>GGI"J"LLNOQxQxSSU\V\XXZ@[@]]_$`$bbdegzgziik^l^nnpBqBssu&v&xxz { }|}|``҄҆BB$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  WW)=*=44>?IIST^r^rhhrRsR} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22zz` `""$F%F''),*,,,./11335k6k88:Q;Q==?3@3BBDEGGIIKhLhNNPJQJSSU,V,XXZ[]]__aabaddfCgCiik%l%nnpqsxsxuuwZxZzz|<}<nnދލNN b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1R} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1R} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1R } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  bb)T*T44>D?DIIS0T0^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppbb T T  FF88**  "#%%'(*y*y,,.k/k113]4]668O9O;;=A>A@@B3C3EEG%H%JJLMOOQRTxTxVVXfYf[[]T^T``b@c@eeg*h*jjlmooqqsmtmvvxSyS{{}9~9uuYYˑ˓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2qX2`qX2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  XX)B*B44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++ w w""$a%a'')K*K,,.5/51134668 9 ;~;~==?h@hBBDRERGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2gW2gW2W2W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIST^^hispsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdaeaggiGjGlln+o+qqs t v~v~xxz`{`}}BB$$vvXX b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2QX2`QX2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d< b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1r  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  VV)@*@44>(?(IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //{{e e""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikmlmnnpQqQssu5v5xxz{}}ooSSƉƋ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2=NY2L =NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)L*L44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$t%t'')d*d,,.T/T113D4D668494;;=$>$@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$T%T'')@*@,,.,/,11346689;z;z==?f@fBBDRERGGI%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2oX2oX2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1O} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1O} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1O } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2׎W2׎W2uW2uW2W2W2W2W2GX2pGX2[X2[X2{bX2{bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vL b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2[W2[W2W2W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2׎W2׎W2{W2{W2W2W2W2W2MX2pMX2[X2[X2{bX2{bX2]X2`]X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vN b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vN b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vN b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2UW2UW2 W2 W2W2W2'W2'W2X2pX2y[X2y[X2_X2_X2X2`X2QX2QX2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDlElGGITJTLLN?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?G@GBBD-E-GGIJLLNNPlQlSSURVRXXZ8[8]]_`bbdeggiikglgnnpMqMssu3v3xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W27GW27GW2ՇW2 ՇW2UW2UW2 W2 W2W2W2'W2'W2X2pX2y[X2y[X2_X2_X2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44>?IIST^x^xhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2oO\2T oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]n^nhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?'@'BBD E GzGzIIK\L\NNP>Q>SSU V XXZ[]s]s__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W21GW21GW2χW2 χW2OW2OW2W2W2W2W2!W2!W2X2pX2s[X2s[X2_X2_X2eX2`eX2X2X2X2X21NY2L 1NY2#Y2 #Y2KZ2p KZ2)[2t )[2}P\2T }P\2s ]2s ]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%}%}'')o*o,,.a/a113S4S668E9E;;=7>7@@B)C)EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m ooqqsotovvxUyU{{};~;!!ww[[͓͑ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2MW2 MW2ˎW2ˎW2W2W2W2W2W2W2SX2pSX2[X2[X2obX2obX2X2`X2!X2!X2X2X21NY2T 1NY2Y2 Y29Z2x 9Z2[2| [2P\2\ P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1\  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  LL)@*@44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //  k k""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`MbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%%mm b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W21GW21GW2χW2 χW2OW2OW2W2W2W2W2!W2!W2X2pX2m[X2m[X2_X2_X2X2`X2X2X2X2X21NY2L 1NY2 Y2  Y23Z2p 3Z2[2t [2 P\2T  P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)B*B44>.?.IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln/o/qqstvvxxzd{d}}FF(( zz\\ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2SW2 SW2ˎW2ˎW2W2W2W2W2W2W2SX2pSX2[X2[X2obX2obX2kX2`kX2X2X2X2X21NY2T 1NY2Y2 Y2-Z2x -Z2 [2|  [2O\2\ O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  JJ)<*<44>$?$IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD4?4IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : : qq__{{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjglln\o\qqs@t@vvx$y${{}~zz^^ЇЉBB&&# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2eW2deW2W2lW2W2W2ӛW2ӛW2W2@W27W27W2W2W2oX2@oX2X2 X2kX2kX2 \X2 \X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2X2 X2 X2H  X2MX2hMX29Y2D9Y2;Y2;Y2]Y2]Y2hY2 hY2Y2Y2Z2pZ2/[2t/[2[2T[2P\2P\2Y"]2$Y"]25]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K[[  &  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dKa,wB X#Gn9E  O   e 0  ; ?  Q bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)*44>5?5IIS!T!^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$m%m'')]*],,.M/M11346688:w;w=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjsllnkokqqsWtWvvxCyC{{}/~/}}iiߑߓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W23W23W2W2|W2W2HW2ǰW2ǰW2eW2PeW2W2XW2W2W2X2PX2X2X2X2X27[X2$ 7[X2`X2, `X2cX2 cX2amX2$ amX2pX2 pX2ٝX2| ٝX2X2XX2X2X2=X2=X21 Y21 Y2=Y2\=Y2^Y2^Y2iY2iY2Y28Y2Z2Z2[2[2[2[2O\2lO\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O| b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O| b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O|  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1dH   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1ddH,tX< dF  @ (   >  l  N  0z 8x| bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ HH):*:44>&?&IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^HH 2 2 x xbbLLeeO O""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnoqtqtssuXvXxxz<{<}} vvZZ̎̐>># b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2kW2dkW2W2lW2W2W2ӛW2ӛW2W2@W27W27W2W2W2oX2@oX2X2 X2kX2kX2 \X2 \X2bX2 bX2AlX2 AlX2woX2d woX2X2 X2;X2 ;X2X2H X2X2hX29Y2D9Y2;Y2;Y2]Y2]Y2hY2 hY2 Y2 Y2Z2pZ2[2t[2[2T[2P\2P\2!]2$!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ka  &  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dKa,wB X#n9   O D    e B 0  ;  ~ Q bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)*44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$O%O'');*;,,.'/'1n1n335Z6Z88:F;F==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W21GW21GW2χW2dχW2W2lW2W2W2ٛW2ٛW2W2@W2W2W2W2W2uX2@uX2X2 X2X2X2s[X2s[X2bX2 bX2GlX2 GlX2}oX2d }oX2X2 X2X2 X2aX2H aX2X2hX2QY2DQY2 .?.ItItST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22&&  "#%%'')p*p,,.\/\113H4H668494;;= > @@B C EEGGInJnLLNVOVQQS>T>VVXXZl[l]]_T`TbbdjejggiRjRllnDoDqqs,t,vvxy{{}}ppXẊ̉@@((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2_W2d_W2W2lW2W2W2ٛW2ٛW2W2@W21W21W2W2W2uX2@uX2X2 X2eX2eX2\X2\X2bX2 bX2GlX2 GlX2}oX2d }oX2X2 X2GX2 GX2X2H X2X2hX2O Y2DO Y2 >Y2 >Y2]Y2]Y2hY2 hY2 Y2 Y2Z2pZ2[2t[2[2T[2AP\2AP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K5/o o  : b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dKa,wB Xk#(ne9"  O   e 0  ;  Q bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hhrxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln2o2qqstvvxy{M{M}}77!!ćĉ99##  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W21GW21GW2χW2dχW2W2lW2W2W2ٛW2ٛW2W2@W2W2W2W2W2uX2@uX2X2 X2X2X2m[X2m[X2bX2 bX2GlX2 GlX2}oX2d }oX2X2 X2X2 X2CX2H CX2X2hX2 Y2D Y2>Y2>Y2]Y2]Y2hY2 hY2Y2Y2Z2pZ2 [2t [2[2T[2O\2O\2!]2$!]2G4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 WK bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 WK bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 WK  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 c5 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K:zA F   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dKa,wBZ%p< i  R &   c h 3  >  T T bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR):*:44>"?"IeIeST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$f%f'')N*N,,.6/611346689;z;z==?b@bBBDJEJGGI2J2LLNOQQSSU=V=XXZ![!]]_`bbdeggiiknlnnnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2eW2deW2W2lW2W2W2ٛW2ٛW2W2@W21W21W2W2W2uX2@uX2X2 X2eX2eX2\X2\X2bX2 bX2GlX2 GlX2}oX2d }oX2X2 X2AX2 AX2X2H X2X2hX2Y2DY2C"?"IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD?IIST^u^uhhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$C%C'')-*-,,./11346v6v88:`;`==?H@HBBD.E.GGIJLLNNPmQmSSUSVSXXZ9[9]]_`bbdegxgxiik^l^nnpDqDssu*v*xxz{}}ii܄܆OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2W2TW2W2W2ߛW2ߛW2W2(W2%W2%W2W2W2{X2({X2X2X2YX2YX2[X2[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2X2 X2gX20 gX2X2PX2eY2,eY2[Y2[Y2 Y2$ Y2Z2Z2[2[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 47W b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18}B n  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2W2TW2W2W2ߛW2ߛW2W2(W2W2W2W2W2{X2({X2X2X2X2X2s[X2s[X2bX2 bX2MlX2 MlX2oX2L oX2ŜX2 ŜX2GX2 GX2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2cO\2cO\23!]23!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UK bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UK bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UK  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X  T  F  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  JJ)0*044>?IISS]g^ghhrIsI} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%mmS S""$9%9'')*,,./1x1x335^6^88:D;D==?(@(BBD E G{G{IIK]L]NNP?Q?SSU!V!XXZ[]t]t__aVbVddf8g8iiklnnpprmsmuuwOxOzz|1}1eeՋՍEE b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2%W2T%W2W2W2W2W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2ǦX2 ǦX2'X20 'X2iX2PiX2Y2,Y2k\Y2k\Y2Y2$Y2Z2Z2[2[2O\2O\2{!]2{!]2#4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o. b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1L  V  3 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2X20 X2_X2P_X2Y2,Y2 \Y2 \Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  MM  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  MM  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  MM   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18J  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>$?$IIS T ^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hbbd0e0ggijllnoqtqtssu\v\xxzD{D}},,tt^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X2WX2PWX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2/[2/[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1['  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1<|  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`0 4t } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W27GW27GW2ՇW2LՇW2+W2T+W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2;X2 ;X2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  LL bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  LL bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  LL  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ccmh b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\ % n 7 W } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>!?!IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikflfnnpJqJssu.v.xxz{}}hhڄچNN44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2=W2T=W2ǘW2ǘW2W2W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2˥X2 ˥X2+X20 +X2mX2PmX2'Y2,'Y2C]Y2C]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t8 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 P p @  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2SX20 SX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2#[2#[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ic@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ic@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ic@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  @ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations15u  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d8 T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW21W2T1W2W2W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2ץX2 ץX27X20 7X2yX2PyX2Y2,Y2]Y2]Y2#Y2$#Y2Z2Z25[25[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1:z bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hC y < } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$o%o'')_*_,,.O/O113?4?668/9/;;=>@@BCEEGGIwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2YX2 YX2X20 X2X2PX2'Y2,'Y2m]Y2m]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jb> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jb> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jb>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ` 4  \ ( } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2}X20 }X2X2PX2Y2,Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y"BBb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T / e  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2MX2 MX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 BK bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 BK bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 BK  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hG&  Z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  ZZ)F*F44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln>o>qqs&t&vvxy{{}}jjނބRRƇƉ::"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2+W2T+W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2[X20 [X2X2PX2Y2,Y2E^Y2E^Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mE  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ? b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations11q bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@& r  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  UU)C*C44>1?1IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd`e`ggiJjJlln4o4qqstvvxy{}{}}}gg܂܄QQƇƉ;;%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2_X2 _X2X20 X2X2PX2?Y2,?Y2?^Y2?^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 AM bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 AM bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 AM  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 !  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d C "   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2+W2T+W2W2W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2)X2 )X2X20 X2X2PX27Y2,7Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~,LLl b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 X > $ f } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X23X2P3X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations17w  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1kN 1 b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W2+W2T+W2W2W2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2YX2 YX2X20 X2X2PX2Y2,Y2o^Y2o^Y2Y2$Y2Z2Z2/[2/[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b:  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 { b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/o  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P8  $ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99 yy b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2YX2 YX2X20 X2X2PX2?Y2,?Y2^Y2^Y2Y2$Y2Z2Z2[2[2{O\2{O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |x(H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  w Z = E } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d>   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 7<\ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 p X @ ( } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  UU)=*=44>#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbdegsgsiikWlWnnp;q;ssuvxxz{}u}uYY˄ˆ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW21W2T1W2W2W2W2W2!W2(!W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2CX2PCX2mY2,mY2^Y2^Y2Y2$Y2Z2Z2/[2/[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 6 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 D  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`N< * (h } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW25W2L5W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2YX2 YX2X20 X2X2PX2Y2,Y2;_Y2;_Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1j bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  ]])O*O44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk]] O O  AA33%%  " # %%'')t*t,,.f/f113X4X668J9J;;=<><@@B.C.EEG H JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2UX20 UX2X2PX2 Y2, Y2e_Y2e_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @L bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @L bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @L  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 5 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  t b } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W21W2T1W2W2W2W2W2iW2(iW2W2W2W2W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2wX2 wX2X20 X2X2PX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2/[2/[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1]  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1t g Z M] } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>3?3IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!uuWW b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2aW2$aW29W29W2W2W23W2$3W2 X2 X2[X2[X2`X2$`X2X2X2X2X2-X2 -X2ONY2| ONY2;Y2 ;Y2Z2 Z2M[2 M[2eP\2 eP\25"]25"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>A?AIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIiJiLLNWOWQQSETEVVX3Y3[[]!^!``bceeggitjtlln]o]qqsAtAvvx%y%{{} ~ {{__чщCC'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2-X2-X2ONY2L ONY2GY2 GY2Z2p Z2Y[2t Y[2P\2T P\2}"]2}"]2%5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)T*T44>B?BIIS.T.^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44$$  "#%|%|'')l*l,,.\/\113L4L668<9<;;=,>,@@BCEEG H JJLLNnOnQQSZTZVVXFYF[[]2^2``bceeg h jjllnlolqqsXtXvvxDyD{{}0~0~~jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2W2aW2$aW23W23W2W2W23W2$3W2X2X2[X2[X2`X2$`X2?X2?X2X2X2-X2 -X2ONY2| ONY2)Y2 )Y2Z2 Z2;[2 ;[2 P\2  P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  __)I*I44>3?3IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uu__ I I  33||f f""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUtVtXXZ^[^]]_H`Hbbd2e2ggijllnoququssuYvYxxz={=}}!!ww[[͎͐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2cW2cW2W2W2W2W25X2p5X2[X2[X2bX2bX2X2`X2X2X2-X2-X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K% b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)L*L44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$^%^'')J*J,,.6/6113"4"6689;;==?p@pBBD\E\GGIFJFLLN.O.QQSTVVXXZr[r]]_Z`ZbbdBeBggi*j*llnoqqssunvnxxzV{V}}>>&&jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2[W2[W23W23W2W2W2-W2-W2X2pX2[X2[X2_X2_X2-X2`-X2X2X2'X2'X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  kk)W*W44>A?AIIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG;;//  "###%%'(**,,.q/q113]4]668I9I;;=5>5@@B!C!EEG H JJLLNiOiQQSQTQVVX9Y9[[]!^!``b c e}e}ggiejellnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>8?8IIS"T"^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGI`J`LLNJOJQQS4T4VVXY[[]^`}`}bbdgegggiQjQlln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)M*M44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%y%y'')a*a,,.I/I1131416689;;=>@u@uBBD]E]GGICJCLLN'O'QQS T V}V}XXZa[a]]_E`Ebbd)e)ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2WX2WX2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T. b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])G*G44>/?/IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd$e$ggi j l}l}nnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2aW2aW2-W2-W2W2W23W23W2X2pX2[X2[X2`X2`X2 X2` X2iX2iX2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1&F  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>+?+IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?W@WBBD=E=GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikmlmnnpSqSssu9v9xxz{}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2SX2`SX2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]5 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)L*L44>4?4IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@00  "#%~%~'')j*j,,.V/V113B4B668.9.;;=>@@BCEtEtGGI\J\LLNDODQQS,T,VVXY[[]]_p`pbbdXeXggi@j@lln(o(qqstvvxxzl{l}}TTȂȄ<<!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  YY)?*?44>!?!IIST^t^thhrTsT} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    ||b b""$H%H'').*.,,./11335m6m88:S;S==?5@5BBDEGGIIKjLjNNPLQLSSU.V.XXZ[]]__acbcddfEgEiik'l'nnp q szszuuw\x\zz|>}> ppPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2WX2`WX2X2X2!X2!X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]7 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)D*D44>(?(IIS T ^~^~hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00 x x""$`%`'')H*H,,.0/011346689;t;t==?X@XBBD[>]]_"`"bbdegxgxiik\l\nnp@q@ssu$v$xxz{}z}z^^ЄІ@@"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  \\)F*F44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// { {""$e%e'')O*O,,.9/9113#4#668 9 ;;==?l@lBBDVEVGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77}}gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ^^)P*P44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll^^ P P  BB44&&  " # %%'')u*u,,.g/g113Y4Y668K9K;;==>=@@B/C/EEG!H!JJLMOOQQStTtVVXbYb[[]P^P``b*?*IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUȉȋ;;!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2cX2`cX2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ZZ)H*H44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llZZ H H  66$$ w w""$e%e'')S*S,,.A/A113/4/6689;;= > @@BBDpEpGGI\J\LLNFOFQQS0T0VVXY[[]^`y`ybbd_e_ggiEjElln)o)qqs t v|v|xxz^{^}}@@""ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2EX2`EX2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oG b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2aW2aW2W2W2W2W23W23W2X2pX2[X2[X2`X2`X2ӣX2`ӣX23X23X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  SS)=*=44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiSS = =  ''ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?K@KBBD5E5GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xP b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xR b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xR b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xR b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Dd  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>?IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2׎W2׎W2W2W2W2W2W2W2SX2pSX2[X2[X2{bX2{bX2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2#[2t #[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y% b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>(?(IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;= > @@BBDhEhGGIPJPLLN8O8QQS T VVXY[|[|]]_d`dbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2UW2UW2 W2 W2W2W2'W2'W2X2pX2y[X2y[X2_X2_X2 X2` X2iX2iX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2iO\2T iO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2׎W2׎W2W2W2W2W2W2W2SX2pSX2[X2[X2{bX2{bX2X2`X2KX2KX2X2X27NY2L 7NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?L@LBBD0E0GGIJLLNNPjQjSSUNVNXXZ2[2]]_`bbdegtgtiikXlXnnp%?%IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2SW2 SW2юW2юW2W2W2%W2%W2W2W2YX2pYX2[X2[X2ubX2ubX2kX2`kX2X2X2X2X27NY2T 7NY2#Y2 #Y2KZ2x KZ2)[2| )[2P\2\ P\2y ]2y ]2!3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b. b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  RR)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``RR D D  66((   ""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W27GW27GW2ՇW2 ՇW2IW2IW2W2W2W2W2W2W2X2pX2s[X2s[X2_X2_X2X2`X2!X2!X2X2X27NY2L 7NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]2s2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL):*:44>"?"IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2SW2 SW2юW2юW2W2W2%W2%W2W2W2YX2pYX2[X2[X2ubX2ubX2#X2`#X2X2X2X2X27NY2T 7NY2Y2 Y29Z2x 9Z2[2| [2 P\2\  P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)D*D44>0?0IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2FW2FW2W2|W2W2W23W23W2W2|W2W2HW2ͰW2ͰW2kW2PkW2W2XW2W2W2X2PX2X2X2X2X2=[X2$ =[X2`X2, `X2cX2 cX2amX2$ amX2pX2 pX2ٝX2| ٝX2%X2X%X2X2X2X2X2?Y2?Y2;Y2\;Y2=]Y2=]Y2;hY2;hY2#Y28#Y2Z2Z25[25[2[2[2}P\2l}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |&F\| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1HHb  /  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dMAzg34Nh5O  i  6   P  ]  7w bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` RR)N*N44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``RR D D qq   ""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')u*u,,.a/a1134668 9 ;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWllnKoKqqs3t3vvxy{{}~ww__ӇӉGG//  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W23W23W2W2|W2W2HW2ͰW2ͰW2kW2PkW2W2XW2W2W2X2PX2X2X2X2X2=[X2$ =[X2`X2, `X2cX2 cX2amX2$ amX2pX2 pX2ٝX2| ٝX2 X2X X2iX2iX2X2X2y Y2y Y25>Y2\5>Y2_Y2_Y2jY2jY2 Y28 Y2Z2Z2[2[2[2[2 P\2l P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 z$>^ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1o z ' g b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1do^M<l%+` xeR  ?  ,    W   s `O>X bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` NN)D*D44>2?2IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < < pp^^}}k k""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln3o3qqstvvxxzh{h}}JJ,,bb# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2qW2dqW2W2lW2W2W2ٛW2ٛW2W2@W2=W2=W2W2W2uX2@uX2X2 X2qX2qX2\X2\X2bX2 bX2GlX2 GlX2}oX2d }oX2X2 X2X2 X2X2H X2;X2h;X2)Y2D)Y2'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')S*S,,.;/;1~1~335f6f88:N;N==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II--  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W27GW27GW2ՇW2dՇW2W2lW2W2W2ߛW2ߛW2 W2@ W2W2W2W2W2{X2@{X2X2 X2X2X2y[X2y[X2bX2 bX2MlX2 MlX2oX2d oX2ŜX2 ŜX2ťX2 ťX2%X2H %X2gX2hgX2KY2DKY20?0IvIvST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``PP @ @  00   x x""$h%h'')X*X,,.H/H113848668(9(;;=>@@BCEEGGIpJpLLN\O\QQSHTHVVXY[z[z]]_f`fb c eegginjnllndodqqsPtPvvxp{h8`XPH@:4 . ( "   Hz4t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIST^J^Jhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbd e g}g}iikclcnnpqssuuwrxrz{}}nnTTƉƋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W27GW27GW2ՇW2dՇW2W2lW2W2W2ߛW2ߛW2W2@W2W2W2W2W2{X2@{X2X2 X2X2X2s[X2s[X2bX2 bX2MlX2 MlX2oX2d oX2ŜX2 ŜX2ťX2 ťX2%X2H %X2gX2hgX2EY2DEY2$?$IgIgS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$J%J'')6*6,,."/"11346688:p;p==?\@\BBDHEHGGI4J4LLNOQQSTVGVGXXZ/[/]]_`bbd0e0ggijllnoq|q|ssudvdxxzL{L}}44xx``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2eW2deW2W2lW2W2W2ߛW2ߛW2W2@W27W27W2W2W2{X2@{X2X2 X2kX2kX2 \X2 \X2bX2 bX2MlX2 MlX2oX2d oX2ŜX2 ŜX2 X2  X2iX2H iX2X2hX2Y2DY2=?IISS];^;hhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__a[b[ddf=g=iikln_n_pprAsAuuw#x#zz|6}6jjڋڍJJ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2#X2 #X2X20 X2X2PX2X2,X2OZY2OZY2 Y2$ Y2Z2Z2[2[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TM bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TM bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TM  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2RRr b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations15u bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1A~;   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)<*<44>*?*IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$Y%Y'')G*G,,.5/5113#4#6689;;==?v@vBBDdEdGGIRJRLLN'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2 W2( W2W2W2W2W2X2(X2X2X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2AX2 AX2X20 X2X2PX2Y2,Y2ZY2ZY2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UM bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UM bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UM  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |Oot b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x5r/Q  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)7*744>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz``ӉӋFF,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2%W2T%W2W2W2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X2CX2PCX2X2,X2c[Y2c[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n:  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Rr b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1c*q '  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2_X2 _X2X20 X2X2PX2[X2,[X2ZY2ZY2 Y2$ Y2Z2Z2[2[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IO  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IO  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IO   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1DLP  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;=>@@BBDjEjGGIRJRLLN:O:QQS"T"VVX Y [~[~]]_f`fbbdNeNggi6j6llnoqqstvzvzxxzb{b}}JJ22rr b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2=W2T=W2ǘW2ǘW2W2W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2[X2,[X2[Y2[Y2Y2$Y2Z2Z2#[2#[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x @@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R$v P  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2AX2 AX2X20 X2X2PX2X2,X2Q[Y2Q[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HN bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HN bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HN  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T\   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^t^thhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbddfngniikRlRnnp6q6ssuvxxzz|p}pTTƄƆ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2=W2T=W2ǘW2ǘW2W2W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2X20 X2=X2P=X2Y2,Y2#\Y2#\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Hl>  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  MM)7*744>#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2iY2,iY2[Y2[Y2#Y2$#Y2Z2Z25[25[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ie@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ie@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ie@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Om @  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)H*H44>:?:IIS&T&^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%{%{'')m*m,,._/_113Q4Q668C9C;;=5>5@@B'C'EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m o}o}qqsctcvvxIyI{{}/~/mmߌߎQQÑÓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2%X2P%X2[X2,[X2A\Y2A\Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {&F b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]:  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  ]])I*I44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii]] Q Q  EE99--!!  "#%%'(*w*w,,.c/c113O4O668;9;;;='>'@@BCEEGGIuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2Y2,Y2\Y2\Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jd> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jd> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jd>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1< Z)  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW27W2T7W2W2W2W2W2W2(W2%W2%W2 W2 W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2SX2 SX2X20 X2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Vvv b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tQ. _ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2kX2 kX2X20 X2 X2P X2aX2,aX25\Y25\Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Z4  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)D*D44>4?4IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln^o^qqsJtJvvx6y6{{}"~"pp\\ґғ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW21W2T1W2W2W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2AX2PAX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1oI  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hP8p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?O@OBBD5E5GGIJLLNOQtQtSSUZVZXXZ@[@]]_&`&bbd e ggiikelennpKqKssu1v1xxz{}}ppVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2}X20 }X2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 <Aa b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hBvJ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW21W2T1W2W2W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2/X2 /X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2uO\2uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 1QVv b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 pXX } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  OO)5*544>?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW21W2T1W2W2W2W2W2!W2(!W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2qX2 qX2X20 X2X2PX2cY2,cY2\Y2\Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1< bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fK0a } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW21W2T1W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2qX2 qX2X20 X2X2PX2gX2,gX2]Y2]Y2Y2$Y2Z2Z2/[2/[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d>   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jT> } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  VV)@*@44>+?+IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99)) u u""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33{{ee b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW21W2T1W2W2W2W2W2!W2(!W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2wX20 wX2X2PX2 Y2, Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 7<\ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1z_DF } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW21W2T1W2W2W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2+X2P+X2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(|f:z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>&?&IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W27W2T7W2W2W2W2W2iW2(iW2W2W2 W2 W2X2(X2X2X2;X2;X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2qX2 qX2X20 X2X2PX2gX2,gX2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1uj_@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW27W2T7W2W2W2W2W2!W2(!W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X2IX2PIX2Y2,Y2Q^Y2Q^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 5 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2;W2L;W27W2T7W2W2W2W2W2oW2(oW2 W2 W2 W2 W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2OX20 OX2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  - b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1TI>35 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  QQ);*;44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`M?MIIS9T9^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1yykk ] ]  OOAA33%%  "#%%' ( **,,.t/t113f4f668X9X;;=J>J@@B@?@IIS(T(^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvjj ^ ^  RRFF::..  ""#"%%'(**,,.p/p113\4\668H9H;;=4>4@@B C EEG H JJLLNhOhQQSPTPVVX8Y8[[] ^ ``bce|e|ggidjdllnLoLqqs4t4vvxy{{}~xx``ԇԉHH00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2aW2$aW29W29W2W2W23W2$3W2 X2 X2[X2[X2`X2$`X2EX2EX2X2X2-X2 -X2ONY2| ONY2/Y2 /Y2Z2 Z2A[2 A[2/P\2 /P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS'T'^^h i sysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdjejggiPjPlln4o4qqstvvxxzi{i}}KK--aa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2W2W2iW2iW2W2W2W2W2;X2p;X2[X2[X2bX2bX2X2`X2cX2cX2-X2-X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M) b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations17  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  ff)N*N44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvff V V  FF66&&  "#%z%z'')b*b,,.J/J1132426689;;=>@v@vBBD^E^GGIDJDLLN(O(QQS T V~V~XXZb[b]]_F`Fbbd*e*ggijllnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2[W2[W29W29W2W2W2-W2-W2 X2p X2[X2[X2_X2_X23X2`3X2X2X2'X2'X2ONY2L ONY2AY2 AY2Z2p Z2S[2t S[2P\2T P\2w"]2w"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)U*U44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%}%}'')m*m,,.]/]113M4M668=9=;;=->-@@BCEEG H JJLLNoOoQQS[T[VVXGYG[[]3^3``bceeg h jjllnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2]X2`]X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ZZ)D*D44>*?*IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo[[ G G  33  i i""$S%S'')=*=,,.'/'11346688:p;p==?V@VBBD5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$]%]'')I*I,,.5/5113!4!668 9 ;;==?o@oBBD[E[GGIEJELLN-O-QQSTVVXXZq[q]]_Y`YbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2SX2`SX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2[2t [2O\2T O\2W!]2W!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V2 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ZZ)@*@44>"?"IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  }}c c""$I%I'')/*/,,./11335n6n88:T;T==?6@6BBDEGGIIKkLkNNPMQMSSU/V/XXZ[]]__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2aW2aW23W23W2W2W23W23W2X2pX2[X2[X2`X2`X2'X2`'X2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS#T#^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbdhehggiRjRlln.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn^^ N N  >>.. z z""$d%d'')N*N,,.8/8113"4"668 9 ;;==?k@kBBDUEUGGI>J>LLN&O&QQSTVVXXZj[j]]_R`Rbbd:e:ggi"j"lln o q~q~ssufvfxxzN{N}}66||ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2aW2aW2-W2-W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)F*F44>.?.IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI2J2LLNOQQSSUqVqXXZW[W]]_=`=bbd#e#ggi j l|l|nnpbqbssuHvHxxz.{.}}mmSSƎƐ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2iX2`iX2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_; b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  YY)C*C44>+?+IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  ~~h h""$R%R'')<*<,,.&/&11346688:o;o==?Y@YBBDCECGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggiikplpnnpTqTssu8v8xxz{}}rrVVɉɋ<<"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  aa)M*M44>5?5IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqee Y Y  MMAA11  " # %%'')k*k,,.W/W113C4C668/9/;;=>@@BCEuEuGGI]J]LLNEOEQQS-T-VVXY[[]]_q`qbbdYeYggiAjAlln)o)qqstvvxxzm{m}}UUɂɄ==""{{ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2=X2`=X2X2X2!X2!X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hB b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Zz  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  XX)F*F44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjXX F F  44""u u""$c%c'')Q*Q,,.?/?113-4-6689;;= > @@BBDnEnGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  [[)C*C44>'?'IIS T ^}^}hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// w w""$_%_'')G*G,,.///11346688:s;s==?W@WBBD;E;GGIJLLNOQuQuSSUYVYXXZ=[=]]_!`!bbdegwgwiik[l[nnp?q?ssu#v#xxz{}y}y]]τφ??!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hD b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  TT)>*>44>(?(IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjTT > >  ((qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLN O QQSSUiViXXZS[S]]_=`=bbd'e'ggijllnnpjqjssuNvNxxz2{2}}llމދPPސ44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  __)Q*Q44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm__ Q Q  CC55''  " # %%'')v*v,,.h/h113Z4Z668L9L;;=>>>@@B0C0EEG"H"JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qK b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJtJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2ߣX2`ߣX2?X2?X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2MX2`MX2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1l  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI8J8LLNOQQSTVrVrXXZV[V]]_:`:bbdeggijltltnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX23X2`3X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zT  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Yy  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>?IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdegzgziik`l`nnpFqFssu,v,xxz{}}kkބކOO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2ǣX2`ǣX2'X2'X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1H} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1H} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1H } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)?*?44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2ݎW2ݎW2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2oO\2T oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zV" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zV" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zV" b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]i^ihhrIsI} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3eeՋՍEE b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2oX2oX2X2X27NY2L 7NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2׎W2׎W2W2W2%W2%W2W2W2YX2pYX2[X2[X2{bX2{bX2X2`X2X2X2X2X27NY2L 7NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]) b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>"?"IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI2J2LLNOQQSTVvVvXXZ^[^]]_F`Fbbd8e8ggi j llnoq|q|ssudvdxxzL{L}}44zzdd b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>"?"IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2юW2юW2W2W2%W2%W2W2W2YX2pYX2[X2[X2ubX2ubX2X2`X2EX2EX2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 -  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2=GW2=GW2ۇW2 ۇW2IW2IW2W2W2W2W2W2W2X2pX2[X2[X2_X2_X2kX2`kX2X2X2X2X27NY2L 7NY2Y2 Y29Z2p 9Z2[2t [2/P\2T /P\2% ]2% ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1D} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1D} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1D } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)C*C44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2YW2 YW2юW2юW2W2W2+W2+W2W2W2_X2p_X2[X2[X2ubX2ubX2GX2`GX2X2X2X2X27NY2T 7NY2Y2 Y2EZ2x EZ2#[2| #[2GP\2\ GP\2= ]2= ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  LL)D*D44>2?2IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2=GW2=GW2ۇW2 ۇW2IW2IW2 W2 W2W2W2W2W2X2pX2y[X2y[X2_X2_X2ͣX2`ͣX2-X2-X2X2X27NY2L 7NY2Y2 Y2'Z2p 'Z2[2t [2O\2T O\2]2]2U2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1E} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1E} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1E } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>?IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2YW2 YW2юW2юW2W2W2+W2+W2W2W2_X2p_X2[X2[X2ubX2ubX2/X2`/X2X2X2X2X27NY2T 7NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1z  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  HH):*:44>$?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^HH 2 2  {{eeO O""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2W2W29W29W2ÙW2|ÙW2W2HW2ͰW2ͰW2kW2PkW2W2XW2 W2 W2X2PX2X2X2X2X2=[X2$ =[X2`X2, `X2cX2 cX2gmX2$ gmX2pX2 pX2ߝX2| ߝX2ݥX2XݥX2=X2=X2X2X2KY2KY2+?+IIST^^his}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ? uuee  k k""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`ObbdCeCggi+j+llnoqqssuovoxxzW{W}}??''qq# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2qW2dqW2W2lW2W2W2ߛW2ߛW2W2@W2=W2=W2W2W2{X2@{X2X2 X2qX2qX2\X2\X2bX2 bX2MlX2 MlX2oX2d oX2ŜX2 ŜX2eX2 eX2X2H X2X2hX2] Y2D] Y2?Y2?Y2g]Y2g]Y2ehY2 ehY2Y2Y2Z2pZ2)[2t)[2[2T[2MP\2MP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1V 4 t  H b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dOm< Z)xGce24    R .  ! p ?   ]- bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)*44>.?.IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/91~1~335l6l88:Z;Z=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln=o=qqs't'vvxy{{}}ppZZχωDD..  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2FW2FW2W2|W2W2W29W29W2ÙW2|ÙW2W2HW2ͰW2ͰW2kW2PkW2W2XW2 W2 W2X2PX2X2X2X2X2=[X2$ =[X2`X2, `X2cX2 cX2gmX2$ gmX2pX2 pX2ߝX2| ߝX2X2XX2qX2qX2X2X25Y25Y2&?&IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7 z zffRRooY Y""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD%?%IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$N%N'')8*8,,."/"1d1d335N6N88:8;8==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpeqessuKvKxxz1{1}}ppVVɎɐ<<& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 ' bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2=GW2=GW2ۇW2dۇW2%W2l%W2W2W2W2W2 W2@ W2W2W2W2W2X2@X2X2 X2X2X2y[X2y[X2bX2 bX2SlX2 SlX2oX2d oX2˜X2 ˜X2 X2  X2mX2H mX2X2hX2KY2DKY2+?+IpIpST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGISJSLLNAOAQQS/T/VtVtXXZb[b]]_P`Pbbdpepggi^j^llnQoQqqs5t5vvxy{{}}ooSSŇʼn77  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2kW2dkW2%W2l%W2W2W2W2W2W2@W27W27W2W2W2X2@X2X2 X2kX2kX2 \X2 \X2bX2 bX2SlX2 SlX2oX2d oX2˜X2 ˜X2X2 X2CX2H CX2X2hX2Y2DY2a,?,IIST^X^Xhhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLN?IaIaS T ^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__II 3 3  ||ffP P""$:%:'')$*$,,./11335m6m88:W;W==?A@ABBD+E+GGIJLLNOQvQvSSU-V-XXZ[]]_`bbdeggijl}l}nnpiqissuMvMxxz1{1}}kk݉݋OO33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2kW2dkW2%W2l%W2W2W2W2W2W2@W2=W2=W2W2W2X2@X2X2 X2qX2qX2\X2\X2bX2 bX2SlX2 SlX2oX2d oX2˜X2 ˜X2X2 X2!X2H !X2cX2hcX2Y2DY2?IIST^E^Ehhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbdegrgriikVlVnnpqsysyuuw]x]z{}t}tXXʄʆ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2X2 X2X20 X2%X2P%X2Y2,Y2K[Y2K[Y2 Y2$ Y2Z2Z2[2[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TO bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TO bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TO  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k& b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1-m  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1j'd!  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  WW)C*C44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33''  "#%%'')q*q,,.]/]113I4I668595;;=!>!@@B C EEGGIoJoLLNWOWQQS?T?VVX'Y'[[]^``bbdkekggiSjSlln;o;qqs#t#vvx y {{}}ggۂۄOOÇÉ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2+W2T+W2W2W2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2ǦX2 ǦX2'X20 'X2iX2PiX2qY2,qY2G\Y2G\Y2Y2$Y2Z2Z2/[2/[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n:  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Pp b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1X bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Z%p;  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2=GW2=GW2ۇW2LۇW2%W2T%W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2SlX2 SlX2oX2L oX2˜X2 ˜X2GX2 GX2X20 X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UO bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UO bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UO  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tty b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^  X  I  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS);*;44>#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$g%g'')O*O,,.7/711346689;{;{==?c@cBBDKEKGGI3J3LLNOQQSSUmVmXXZQ[Q]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2+W2T+W2W2W2W2W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2OX20 OX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 : b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  Q  g S } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2YX2 YX2X20 X2X2PX2eY2,eY2[Y2[Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 EQ  bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 EQ  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 EQ   bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1H bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 f,rX  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  KK)5*544>?IIST^v^vhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?I@IBBD/E/GGIJLLNNPnQnSSUTVTXXZ:[:]]_ ` bbdegygyiik_l_nnpEqEssu+v+xxz{}}jj݄݆PP‰‹44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2=W2T=W2ǘW2ǘW2W2W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2uY2,uY2\Y2\Y2#Y2$#Y2Z2Z25[25[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Rrr b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1[  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10X, | } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2_X2 _X2X20 X2X2PX2Y2,Y2_\Y2_\Y2Y2$Y2Z2Z2[2[2iO\2iO\29!]29!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 DP bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 DP bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 DP  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 ~ D  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]h^hhhrJsJ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__aWbWddf9g9iiklnnpprnsnuuwPxPzz|2}2ff֋֍FF b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2=W2T=W2ǘW2ǘW2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X2#X2 #X2X20 X2X2PX2 Y2, Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .NNn b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 T ( | P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2Y2,Y2\Y2\Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ig@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ig@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ig@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1( h  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xIk ^ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  PP):*:44>%?%IIS T ^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--uu__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW21W2T1W2W2W2W2W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2MX2 MX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ttt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1gF % B } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2)X2 )X2X20 X2X2PX2 Y2, Y21]Y21]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jf> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jf> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jf>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0PUu b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 < ^ /  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW27W2T7W2W2W2W2W2W2(W2%W2%W2 W2 W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X2X2PX2AY2,AY2c^Y2c^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\(  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 %EEe b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1!  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  c B ! ! } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW27W2T7W2W2W2W2W2!W2(!W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2_X20 _X2X2PX2Y2,Y2m]Y2m]Y2Y2$Y2Z2Z2)[2)[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  z%E b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations14t bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P, d $ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>/?/IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnKoKqqs/t/vvxy{{}}iiۂۄMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X2YX2PYX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2/[2/[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1qM  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations11q  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`J4  H } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)F*F44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::**  "#%x%x'')d*d,,.P/P113<4<668(9(;;=>@@BBDpEpGGIXJXLLN@O@QQS(T(VVXY[[]]_l`lbbdTeTggi#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2#X2 #X2X20 X2X2PX2?Y2,?Y2;_Y2;_Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 &FKk b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 r \ F } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW27W2T7W2W2W2W2W2'W2('W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2qX2 qX2X20 X2X2PX2Y2,Y29^Y29^Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1)w ^  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW27W2T7W2W2W2W2W2uW2(uW2W2W2 W2 W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2qX20 qX2X2PX2Y2,Y2Y_Y2Y_Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fB  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ,1Q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1n bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18- "   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW27W2T7W2W2W2W2W2{W2({W2W2W2 W2 W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2YX2 YX2X20 X2X2PX2Y2,Y2 `Y2 `Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 | b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1#c  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  v k } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln'o'qqs t vzvzxxz\{\}}>> ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2X20 X2-X2P-X2Y2,Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 4T b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1` bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rd V p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2=W2T=W2ǘW2ǘW2W2W2oW2(oW2 W2 W2W2W2X2(X2X2X2AX2AX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2qX2 qX2X20 X2X2PX2Y2,Y2%`Y2%`Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[7  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$r%r'')b*b,,.R/R113B4B668292;;=">"@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2=W2T=W2ǘW2ǘW2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2{Y2,{Y2`Y2`Y2Y2$Y2Z2Z2/[2/[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1];  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 R  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)@*@44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88%%  ""$n%n'')X*X,,.B/B113,4,6689;;=>@u@uBBD_E_GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((pp b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2 X2 X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>;?;IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIcJcLLNMOMQQS7T7VVX!Y![[] ^ ``bbdjejggiTjTlln>o>qqs(t(vvxy{{}}qq[[ЇЉEE// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2 W2aW2$aW2?W2?W2W2W23W2$3W2X2X2[X2[X2`X2$`X2KX2KX2X2X2-X2 -X2UNY2| UNY2#Y2 #Y2Z2 Z25[2 5[2O\2 O\2!]2!]2q4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  aa)K*K44>3?3IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd"e"ggijlxlxnnp\q\ssu@v@xxz${$}}zz^^щыDD** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2 ;W2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2 X2 X2-X2-X2UNY2L UNY2)Y2 )Y2Z2p Z2;[2t ;[2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)J*J44>2?2IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI6J6LLNOQQSTVuVuXXZ[[[]]_A`Abbd'e'ggi j llnnpfqfssuLvLxxz2{2}}qqWWʎʐ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2 X2 X2'X2'X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>?>IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttbb P P  >>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIfJfLLNTOTQQSBTBVVX0Y0[[]^``b c eeggiqjqllnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2 X2 X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)O*O44>7?7IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC33  " # %%'')m*m,,.Y/Y113E4E668191;;=>@@BCEwEwGGI_J_LLNGOGQQS/T/VVXY[[]]_s`sbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2[W2[W29W29W2W2W2-W2-W2 X2p X2[X2[X2_X2_X2KX2`KX2X2X2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1^~  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  \\)F*F44>0?0IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr\\ F F  00yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUqVqXXZ[[[]]_E`Ebbd/e/ggijllnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2oX2`oX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X6 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)G*G44>+?+IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33 { {""$c%c'')K*K,,.3/311346689;w;w==?[@[BBD?E?GGI#J#LLNOQyQySSU]V]XXZA[A]]_%`%bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2aW2aW23W23W2W2W23W23W2X2pX2[X2[X2`X2`X23X2`3X2X2X2'X2'X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Jj  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  hh)T*T44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD88,,  " # %%' ( **,,.n/n113Z4Z668F9F;;=2>2@@BCEEG H J~J~LLNfOfQQSNTNVVX6Y6[[]^``bcezezggibjbllnJoJqqs2t2vvxy{{}~vv^^҇҉FF.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2X2`X2 X2 X2'X2'X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  aa)S*S44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ooaa S S  EE77))  " # %%'')x*x,,.j/j113\4\668N9N;;=@>@@@B2C2EEG$H$JJLMOOQRTwTwVVXeYe[[]S^S``b?c?eeg)h)jjlmooqqsltlvvxRyR{{}8~8ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2aW2aW23W23W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)J*J44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%v%v'')^*^,,.F/F113.4.6689;;==?r@rBBDZEZGGI@J@LLN$O$QQSTVzVzXXZ^[^]]_B`Bbbd&e&ggi j l|l|nnp`q`ssuDvDxxz({(}} ~~bbԉԋFF** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2oX2`oX2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a?  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a?  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a?  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ]])K*K44>7?7IIST^^hisqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo]] K K  99'' z z""$h%h'')V*V,,.D/D113242668 9 ;;=>@@BBDsEsGGI_J_LLNIOIQQS3T3VVXY[[]^`|`|bbdbebggiHjHlln,o,qqstvvxxza{a}}CC%%wwYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2{X2{X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1F} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1F} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1F } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations17W  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  XX)B*B44>(?(IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?T@TBBD:E:GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikjljnnpPqPssu6v6xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2QX2`QX2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jF b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)K*K44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++   ""$s%s'')c*c,,.S/S113C4C668393;;=#>#@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllncocqqsOtOvvx;y;{{}'~'uuaaבד b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  VV)<*<44>?IIST^q^qhhrQsQ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11yy_ _""$E%E'')+*+,,./11335j6j88:P;P==?2@2BBDEGGIIKgLgNNPIQISSU+V+XXZ [ ]~]~__a`b`ddfBgBiik$l$nnpqswswuuwYxYzz|;};mm݋ݍMM b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jH b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //}}i i""$U%U'')A*A,,.-/-11346689;{;{==?g@gBBDSESGGI=J=LLN%O%QQS T VVXXZi[i]]_Q`Qbbd9e9ggi!j!lln o q}q}ssuevexxzM{M}}55yyaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  YY)C*C44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, x x""$b%b'')L*L,,.6/6113 4 668 9 ;;==?i@iBBDSESGGI/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2aW2aW2!W2!W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  UU)?*?44>'?'IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2׎W2׎W2W2W2W2W2W2W2SX2pSX2[X2[X2{bX2{bX2iX2`iX2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)>*>44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2OW2OW2!W2!W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|X$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33qq b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2OW2OW2W2W2W2W2!W2!W2X2pX2[X2[X2_X2_X2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1D} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1D} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1D } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2 [2t  [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|Z& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|Z& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|Z& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>?IIST^u^uhhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2oX2oX2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16V  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>2?2IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2юW2юW2W2W2+W2+W2W2W2_X2p_X2[X2[X2ubX2ubX2X2`X2X2X2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMO}O}QQSkTkVVXYYY[[]G^G``b3c3eeghjjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2CGW2CGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>$?$IIST^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2юW2юW2W2W2+W2+W2W2W2_X2p_X2[X2[X2ubX2ubX2٣X2`٣X29X29X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1$  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdVeVggi?IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdeggiiklllnnpRqRssu8v8xxz{}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2_W2 _W2׎W2׎W2W2W2+W2+W2W2W2_X2p_X2[X2[X2{bX2{bX2AX2`AX2X2X2X2X27NY2T 7NY2)Y2 )Y2QZ2x QZ2/[2| /[2P\2\ P\2 ]2 ]233]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j6 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  OO)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //w w""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2=GW2=GW2ۇW2 ۇW2CW2CW2W2W2W2W2W2W2X2pX2[X2[X2_X2: _X2qX2* qX2X2 X2X2r X27NY2J 7NY2Y2" Y2Z2 Z2[2 [2uO\2 uO\2k]2Z k]22]2z } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1A} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1A} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-w* operation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z   reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })zoutputElements*1152zinput0Elements241472zinput1Elements*9216binvocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1A } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)4*444>?IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2_W2 _W2׎W2׎W2W2W21W21W2W2W2eX2peX2\X2\X2{bX2{bX2X2`X2X2X2X2X27NY2T 7NY2Y2 Y29Z2x 9Z2[2| [2P\2\ P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1L  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  KK)?*?44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##qq] ]""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2W2W29W29W2ÙW2|ÙW2W2HW2ӰW2ӰW2qW2PqW2W2XW2 W2 W2X2PX2X2X2X2X2C[X2$ C[X2`X2, `X2cX2 cX2gmX2$ gmX2pX2 pX2ߝX2| ߝX2٦X2X٦X29X29X2{X2{X2Y2Y2=z  X  6   r P ,lfD bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa QQ)G*G44>1?1IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I   ww!!   ""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdaeaggiIjIlln1o1qqstvvxy{u{u}}]]тфEE,,# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2wW2dwW2%W2l%W2W2W2W2W2W2@W2CW2CW2W2W2X2@X2X2 X2wX2wX2\X2\X2bX2 bX2SlX2 SlX2oX2d oX2˜X2 ˜X2X2 X2aX2H aX2X2hX2Y2DY2m ? IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss] ]""$%''))+r,r..01346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbdeggiiknlnnnpTqTssu:v:xxz { }}yy__щыCC''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2W2W29W29W2ÙW2|ÙW2W2HW2ͰW2ͰW2kW2PkW2W2XW2 W2 W2X2PX2X2X2X2X2=[X2$ =[X2`X2, `X2cX2 cX2gmX2$ gmX2pX2 pX2ߝX2| ߝX2X2XX2GX2GX2X2X2 Y2 Y2q>Y2\q>Y2aY2aY2lY2lY2Y28Y2Z2Z2[2[2[2[2O\2lO\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f< b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1  _  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d 8J#(+  .  1  4  7 - m  :=@MP bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa MM)=*=44>#?#IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A ttdd  i i""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbd e ggiikclcnnpGqGssu+v+xxz{}}eeׄ׆II++ # b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2wW2dwW2%W2l%W2W2W2W2W2W2@W2IW2IW2W2W2X2@X2X2 X2}X2}X2\X2\X2bX2 bX2SlX2 SlX2oX2d oX2˜X2 ˜X2X2 X2gX2H gX2X2hX2Y2DY2?IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qqW W""$%'{'{))+a,a..0|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##uuUUŐŒ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2CGW2CGW2W2dW2+W2l+W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2ťX2 ťX2%X2H %X2gX2hgX2oY2DoY2+7?7IzIzS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaSS E E  77))   ""$x%x'')j*j,,.\/\113N4N668@9@;;=2>2@@B$C$EEGHJJLMO}O}QQSkTkVVX%Y%[[]^``b3c3eeghjjlmooqqsjtjvvxPyP{{}6~6ttXXʑʓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2kW2dkW2+W2l+W2W2W2W2W2W2@W2=W2=W2W2W2X2@X2X2 X2qX2qX2\X2\X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2X2 X2X2H X2WX2hWX2Y2DY2I&?&IIST^P^Phhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd4e4ggijllnnpDqDssu,v,xxzH{H}}00xxbb  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2=GW2=GW2ۇW2dۇW2+W2l+W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2X2 X2X2H X2IX2hIX2]Y2D]Y2+?+IkIkST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLN=O=QQS'T'VgVgXXZQ[Q]]_;`;bbdVeVggiY2 >Y2]Y2]Y2hY2 hY2Y2Y2Z2pZ2 [2t [2[2T[2O\2O\2!]2$!]2G4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t1 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q c  4 b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dQ"sDAf[7(Y*{L   n  ?   a j  2T bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>!?!IIS T ^G^Ghhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikjljnnpqssuuwoxoz{}}llބކRRʼnŋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2X2 X2SX20 SX2X2PX2eX2,eX2][Y2][Y2Y2$Y2Z2Z2/[2/[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TQ" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TQ" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TQ"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w> b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]*w  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>1?1IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$i%i'')Y*Y,,.I/I113949668)9);;=>@@B C EEGGIqJqLLN]O]QQSITIVVX5Y5[[]!^!``b c eeggiojolln[o[qqsGtGvvx3y3{{}~ mmYYϑϓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW21W2T1W2W2W2W2W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2ӦX2 ӦX23X20 3X2uX2PuX2UX2,UX2-[Y2-[Y2Y2$Y2Z2Z2#[2#[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 6V! b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ka  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnLoLqqs0t0vvxy{{}}jj܂܄NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2CGW2CGW2W2LW2+W2T+W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2L oX2ќX2 ќX2ݥX2 ݥX2=X20 =X2X2PX2Y2,Y2[Y2[Y2Y2$Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UQ" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UQ" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UQ"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p0& b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D^+m  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)9*944>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$K%K'')7*7,,.#/#11346688:q;q==?]@]BBDIEIGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--qqYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW21W2T1W2W2W2W2W2W2(W2=W2=W2W2W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2IX2PIX2X2,X2[Y2[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 sIi b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,wB ; {  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2YX20 YX2X2PX2aX2,aX2{[Y2{[Y2Y2$Y2Z2Z2#[2#[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 AS$ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 AS$ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 AS$  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x  @ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1G bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1V%t 7  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hhrtst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdZeZggiDjDlln.o.qqstvvxy{w{w}}aaւքKK55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2=W2T=W2ǘW2ǘW2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X21X2 1X2X20 X2X2PX2[X2,[X2[Y2[Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q1 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q1  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9YYy b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1V,P  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGItJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2qX20 qX2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u::Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ca0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW2UW2TUW2ߘW2ߘW2W2W2!W2(!W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2eX2,eX2[Y2[Y2Y2$Y2Z2Z2#[2#[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ii@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ii@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ii@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1S&y@  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)A*A44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$s%s'')_*_,,.K/K113747668#9#;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33ss b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW27W2T7W2W2W2W2W2W2(W2+W2+W2 W2 W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2GX20 GX2X2PX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q% bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q% bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\(  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q%  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  - b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1aB#@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X25X2 5X2X20 X2X2PX2Y2,Y2Y\Y2Y\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jh> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jh> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jh>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[z` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1LrE  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W27W2T7W2W2W2W2W2W2(W2+W2+W2 W2 W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2X20 X25X2P5X2Y2,Y2%]Y2%]Y2Y2$Y2Z2Z2[2[2uO\2uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`,  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1eF' } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf;?;IIS'T'^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeWW I I  ;;--  "#%|%|'')n*n,,.`/`113R4R668D9D;;=6>6@@B(C(EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m o~o~qqsdtdvvxJyJ{{}0~0nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2_X2 _X2X20 X2X2PX2aX2,aX27]Y27]Y2Y2$Y2Z2Z2)[2)[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sQ  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lXD } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>*?*IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22zzdd b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW27W2T7W2W2W2W2W2!W2(!W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2qX2 qX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xV4 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW27W2T7W2W2W2W2W2W2(W2%W2%W2 W2 W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2}X20 }X2X2PX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU!  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w;@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10t\ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2YX2 YX2X20 X2X2PX2_X2,_X21]Y21]Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1: bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1kT= } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>5?5IIS!T!^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjslln_o_qqsKtKvvx7y7{{}#~#qq]]ӑӓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2=W2T=W2ǘW2ǘW2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2gX2,gX2]Y2]Y2#Y2$#Y2Z2Z25[25[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hF  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1wne` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2wX2 wX2X20 X2X2PX2 Y2, Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1&xaK } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2=W2T=W2ǘW2ǘW2W2W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q" bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q" bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q"  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\SJAW } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>&?&IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnnphqhssuLvLxxz0{0}}jj܉܋NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2mX2,mX2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1th\P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2AW2LAW2CW2TCW2͘W2͘W2W2W2uW2(uW2W2W2W2W2X2(X2X2X2GX2GX2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2mX2,mX2^Y2^Y2)Y2$)Y2Z2Z2;[2;[2eP\2eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1];  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 h b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ``)L*L44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<00$$  "#%%'(*z*z,,.f/f113R4R668>9>;;=*>*@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2wX2 wX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PD8,D } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2CW2TCW2͘W2͘W2W2W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2Y2,Y2e_Y2e_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q# bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q# bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q#  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 1i6V b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2W2aW2$aW2KW2KW2W2W23W2$3W2X2X2[X2[X2`X2$`X2KX2KX2X2X2-X2 -X2UNY2| UNY25Y2 5Y2 Z2  Z2G[2 G[2SP\2 SP\2#"]2#"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  jj)V*V44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ::&&  "#%%'')t*t,,.`/`113L4L668898;;=">"@@B C E~E~GGIfJfLLNNONQQS6T6VVXY[[]^`z`zbbdbebggiJjJlln2o2qqstvvxy{v{v}}^^҂҄FF++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2oW2oW2 W2 W2W2W2AX2pAX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ]])G*G44>-?-IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr^^ J J  66""  l l""$V%V'')@*@,,.*/*11346688:s;s==?Y@YBBD?E?GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz^^ЉЋBB&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2W2aW2$aW2EW2EW2W2W23W2$3W2X2X2[X2[X2`X2$`X2YX2YX2X2X2-X2 -X2UNY2| UNY2Y2 Y2Z2 Z2/[2 /[2O\2 O\2!]2!]2S4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1j  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  dd)L*L44>0?0IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88$$   ""$h%h'')P*P,,.8/8113 4 6689;|;|==?`@`BBDDEDGGI(J(LLN O Q~Q~SSUbVbXXZF[F]]_*`*bbdeggiikdldnnpHqHssu,v,xxz{}}ff؄؆HH**  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2 AW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2YX2`YX2X2X2-X2-X2UNY2L UNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2c!]2c!]2 4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1h  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ]])C*C44>%?%IIST^x^xhhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$  f f""$L%L'')2*2,,./11335q6q88:W;W==?9@9BBDEGGIIKnLnNNPPQPSSU2V2XXZ[]]__agbgddfIgIiik+l+nnp q s~s~uuw`x`zz|B}B$$ttTTĐĒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2'X2'X2ONY2L ONY2AY2 AY2Z2p Z2S[2t S[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  hh)Z*Z44>J?JIIS6T6^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvhh Z Z  LL>>00""  "#%%'(**,,.q/q113c4c668U9U;;=G>G@@B9C9EEG+H+JJLMOOQRT~T~VVXlYl[[]Z^Z``bFcFeeg0h0jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2_X2`_X2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1u  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ^^)H*H44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11 } }""$g%g'')Q*Q,,.;/;113%4%6689;;==?n@nBBDXEXGGIAJALLN)O)QQSTVVXXZm[m]]_U`Ubbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2QX2`QX2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdgegggiMjMlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  \\)F*F44>.?.IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  k k""$U%U'')?*?,,.)/)11346688:r;r==?\@\BBDFEFGGI.J.LLNOQQSSUmVmXXZS[S]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2aW2aW29W29W2W2W23W23W2 X2p X2[X2[X2`X2`X29X2`9X2X2X2'X2'X2INY2L INY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2w"]2w"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)R*R44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%z%z'')j*j,,.Z/Z113J4J668:9:;;=*>*@@BCEEG H JJLLNlOlQQSXTXVVXDYD[[]0^0``bceeghj~j~llnjojqqsVtVvvxByB{{}.~.||hhޑޓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2CX2`CX2X2X2'X2'X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Yy  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>7?7IIS%T%^^h i s{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI_J_LLNMOMQQS;T;VVX)Y)[[]^``bce|e|ggijjjllnSoSqqs7t7vvxy{{}}qqUULJlj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2gW2gW29W29W2W2W29W29W2 X2p X2[X2[X2 `X2 `X2X2`X2X2X2'X2'X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Z%Z'')F*F,,.2/21134668 9 ;;==?l@lBBDXEXGGIBJBLLN*O*QQSTVVXXZn[n]]_V`Vbbd>e>ggi&j&llnoqqssujvjxxzR{R}}::"" ~~ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cC b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  WW)A*A44>+?+IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmWW A A  ++tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O QQSSUlVlXXZV[V]]_@`@bbd*e*ggijllnnpmqmssuQvQxxz5{5}}ooSSŎŐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2UW2UW23W23W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2 X2 X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1B} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1B} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1B } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdeeeggiOjOlln9o9qqs#t#vvx y {{}}llVVˇˉ@@** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2W2W2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2WX2`WX2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lJ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  aa)M*M44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==11%%  "#%%'(*{*{,,.g/g113S4S668?9?;;=+>+@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2eX2`eX2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)C*C44>+?+IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$V%V'')@*@,,.*/*11346688:s;s==?]@]BBDGEGGGI/J/LLNOQQSSUnVnXXZT[T]]_:`:bbd e ggijlylynnp_q_ssuEvExxz+{+}}jj݉݋PPÎÐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lL b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  ]])E*E44>+?+IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==--   ""$q%q'')Y*Y,,.A/A113)4)6689;;==?m@mBBDUEUGGI;J;LLNOQQSTVuVuXXZY[Y]]_=`=bbd!e!ggijlwlwnnp[q[ssu?v?xxz#{#}}yy]]ωϋAA%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2aW2aW2-W2-W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2{X2{X2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ^^)J*J44>2?2IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>..  "#%|%|'')h*h,,.T/T113@4@668,9,;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2׎W2׎W2W2W2%W2%W2W2W2YX2pYX2[X2[X2{bX2{bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uS b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uS b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uS b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>!?!IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2aW2aW2'W2'W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  XX)@*@44>$?$IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, t t""$\%\'')D*D,,.,/,11346688:p;p==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2OW2OW2'W2'W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~\( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--ss]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2OW2OW2!W2!W2W2W2!W2!W2X2pX2[X2[X2_X2_X2ߣX2`ߣX2?X2?X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2KX2KX2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~^* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2UW2UW2!W2!W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2oX2oX2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>4?4IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2юW2юW2W2W2+W2+W2W2W2_X2p_X2[X2[X2ubX2ubX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e1 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLNAOAQQS/T/VVXY[[] ^ ``bbdpepggi^j^llnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2UW2UW2W2W2W2W2'W2'W2X2pX2[X2[X2_X2_X2ͣX2`ͣX2-X2-X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>&?&IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2юW2юW2W2W21W21W2W2W2eX2peX2\X2\X2ubX2ubX2ӣX2`ӣX23X23X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>?IIS T ^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQvQvSSU`V`XXZJ[J]]_4`4bbdeggijl}l}nnpiqissuMvMxxz1{1}}kk݉݋OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2IGW2IGW2W2 W2CW2CW2W2W2W2W2W2W2X2pX2[X2[X2_X2_X2;X2`;X2X2X2X2X27NY2T 7NY2Y2 Y2EZ2x EZ2#[2| #[2SP\2\ SP\2I ]2I ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1f  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)@*@44>,?,IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2eW2 eW2׎W2׎W2W2W21W21W2W2W2eX2peX2\X2\X2{bX2{bX2GX2`GX2X2X2X2X27NY2T 7NY2#Y2 #Y2KZ2x KZ2)[2| )[2YP\2\ YP\2O ]2O ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  UU)K*K44>5?5IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11%%  " # %%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2CGW2CGW2W2 W2CW2CW2W2W2W2W2W2W2X2pX2[X2[X2_X2_X2 X2` X2iX2iX2X2X27NY2T 7NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1R  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM);*;44>#?#IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2eW2 eW2ݎW2ݎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2}X2}X2X2X27NY2T 7NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1"g  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2 W2 W2?W2?W2əW2|əW2W2HW2ӰW2ӰW2qW2PqW2W2XW2W2W2X2PX2X2X2X2X2C[X2$ C[X2`X2, `X2cX2 cX2mmX2$ mmX2pX2 pX2X2| X2}X2X}X2X2X2X2X2Y2Y2O$?$IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1]]II 5 5 u uaaMMwwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}ccՉՋGG++# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2wW2dwW2+W2l+W2W2W2W2W2W2@W2IW2IW2W2W2X2@X2X2 X2}X2}X2\X2\X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2CX2 CX2X2H X2X2hX2Y2DY2[-?-IIS!T!^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$?%?'')+*+,,./113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbd]e]ggiEjElln-o-qqstvvxxzq{q}}YY͂̈́AA((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2 W2 W2?W2?W2əW2|əW2W2HW2ӰW2ӰW2qW2PqW2W2XW2W2W2X2PX2X2X2X2X2C[X2$ C[X2`X2, `X2cX2 cX2mmX2$ mmX2pX2 pX2X2| X2'X2X'X2X2X2X2X2Y2Y2K=Y2\K=Y2ebY2ebY2cmY2cmY2Y28Y2Z2Z2[2[2y[2y[2uO\2luO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 rXx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1`.n< A  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d 0@P`p ~  G *  8 Y F T=}bpP^, bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ PP)6*644>?IIST^y^yhhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7 v vbbNNssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?.@.BBDEGGIIKcLcNNPEQESSU'V'XXZ [ ]z]z__ahbhddfJgJiik,l,nnpqssuuwaxazz|C}C%%wwWWǐǒ# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2}W2d}W2+W2l+W2W2W2W2W2W2@W2OW2OW2W2W2X2@X2X2 X2X2X2!\X2!\X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2X2 X2X2H X25X2h5X2Y2DY2!?!IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$ % '')*,|,|..01346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆGG))   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2CGW2CGW2W2dW2+W2l+W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2X2 X2X2H X2[X2h[X2{Y2D{Y27"?"I`I`ST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``PP @ @  00    l l""$V%V'')@*@,,.*/*11346688:s;s==?]@]BBDGEGGGI2J2LLNOQQSTV@V@XXZ([(]]_`bbd8e8ggi j llnoq|q|ssudvdxxzL{L}}44||ff  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2qW2dqW2+W2l+W2W2W2W2W2W2@W2CW2CW2W2W2X2@X2X2 X2wX2wX2\X2\X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2AX2 AX2X2H X2X2hX2iY2DiY2%;?;IIS'T'^f^fhiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeWW I I  ;;--  "#%|%|'')n*n,,.`/`113R4R668D9D;;=6>6@@B(C(EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m oJoJqqs0t0vvxy{{}2~2ppTTƑƓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2CGW2CGW2W2dW2+W2l+W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2X2 X2SX2H SX2X2hX2a Y2Da Y2>Y2>Y2]Y2]Y2hY2 hY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2G4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 GS& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 GS& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 GS&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s#C b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W.   F b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dW.\3 a8 b7   a d 6 /  z  `  Nf#x bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  KK)5*544>?I[I[S T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?K@KBBD5E5GGIJLLNOQxQxSSU'V'XXZ [ ]]__abdeggiiklllnnpPqPssu4v4xxz{}}nnTTljNj::   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2wW2dwW2+W2l+W2W2W2W2W2W2@W2IW2IW2W2W2X2@X2X2 X2}X2}X2\X2\X2bX2 bX2YlX2 YlX2oX2d oX2ќX2 ќX2GX2 GX2X2H X2X2hX2Y2DY2y d 7 w  T  ] 1  0   V<| bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^S^Shhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>llnnp`q`ssuBvBxxz[{[}}==ssUU  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2X2 X2X20 X2[X2P[X2Y2,Y2w\Y2w\Y2 Y2$ Y2Z2Z2[2[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TS& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TS& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TS&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2R b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1&f  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1vC] 8 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)<*<44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$Y%Y'')G*G,,.5/5113#4#6689;;==?v@vBBDdEdGGITJTLLNBOBQQS0T0VVXY[[] ^ ``bbdqeqggi_j_llnHoHqqs,t,vvxy{{}}ff؂؄JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW27W2T7W2W2W2W2W2W2(W2=W2=W2 W2 W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2SX2 SX2X20 X2X2PX2Y2,Y2s]Y2s]Y2Y2$Y2Z2Z2/[2/[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v_ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fA w 4 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>5?5IIS!T!^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjslln_o_qqsKtKvvx7y7{{}#~#qq]]ӑӓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2IGW2IGW2W2LW21W2T1W2W2W2W2W2W2(W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2L oX2לX2 לX2ѥX2 ѥX21X20 1X2sX2PsX2 Y2, Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 US& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 US& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 US&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 * w D   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)4*444> ? IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$;%;'')%*%,,./11335n6n88:X;X==?B@BBBD,E,GGIJLLNOQwQwSSUaVaXXZK[K]]_5`5bbdeggi j l~l~nnpbqbssuFvFxxz*{*}}dd։֋HH,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW27W2T7W2W2W2W2W2W2(W2=W2=W2 W2 W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2qX2 qX2X20 X2X2PX25Y2,5Y2^Y2^Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R -  c  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$M%M'')9*9,,.%/%11346688:s;s==?_@_BBDKEKGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW27W2T7W2W2W2W2W2!W2(!W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2OX20 OX2X2PX2uY2,uY2\Y2\Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @U( bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @U( bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @U(  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 = b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1}Lj S } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  XX)D*D44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44((  "#%%'')r*r,,.^/^113J4J668696;;=">"@@BCEEGGIpJpLLNXOXQQS@T@VVX(Y([[]^``bbdlelggiTjTlln/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2YX20 YX2X2PX2Y2,Y2%]Y2%]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u? b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 9  W &  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>$?$IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpTqTssu8v8xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2CW2TCW2͘W2͘W2W2W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2wX2 wX2X20 X2X2PX2!Y2,!Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o;  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations17  i B h } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}cc։֋II// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2UW2TUW2ߘW2ߘW2W2W2'W2('W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ik@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ik@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ik@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1*j  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mB f } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)6*644>?IIST^w^whhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdegzgziik`l`nnpFqFssu,v,xxz{}}kkބކQQÉË55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2=W2T=W2ǘW2ǘW2W2W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2X20 X2/X2P/X2Y2,Y2^Y2^Y2Y2$Y2Z2Z2)[2)[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S) bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S) bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`,  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S)  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1( n Q  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)E*E44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii]] Q Q  EE99))  "#%w%w'')c*c,,.O/O113;4;668'9';;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbdSeSggi;j;lln#o#qqs t vvxxzg{g}}OOÂÄ77ww b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2OW2TOW2٘W2٘W2W2W2!W2(!W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2!Y2,!Y2]Y2]Y2Y2$Y2Z2Z2[2[2uO\2uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jj> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jj> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jj>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 l A  k ; } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  LL)2*244>?IISS]i^ihhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ooU U""$;%;'')!*!,,./1z1z335`6`88:F;F==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3gg׋׍GG b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2=W2T=W2ǘW2ǘW2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X25X2 5X2X20 X2X2PX2MY2,MY2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9Y^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1-  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 4  z ] -m } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>!?!IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1!a  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pP0  ,l } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  QQ);*;44>&?&IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqvqvssu^v^xxzF{F}}..vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2=W2T=W2ǘW2ǘW2W2W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2X20 X2-X2P-X2Y2,Y2^Y2^Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uU!  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ',L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1n\ J P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  [[)M*M44>???IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii[[ M M  ??11##  "#%%'')r*r,,.d/d113V4V668H9H;;=:>:@@B,C,EEGHJJLMOOQQSsTsVVXaYa[[]O^O``b;c;eeg%h%jjlmooqqshthvvxNyN{{}4~4rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2#X2 #X2X20 X2X2PX2'Y2,'Y2W^Y2W^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 !AwFf b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p P 0    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>#?#IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikhlhnnpLqLssu0v0xxz{}}jj܄܆PPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2=W2T=W2ǘW2ǘW2W2W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2}X2 }X2X20 X2X2PX2WY2,WY2_Y2_Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY%  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 3  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 &  > } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  UU)C*C44>1?1IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2qX20 qX2X2PX2Y2,Y2^Y2^Y2#Y2$#Y2Z2Z25[25[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ).N b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1L  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1J5 4 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnLoLqqs0t0vvxy{{}}jj܂܄NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2CW2TCW2͘W2͘W2W2W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2YX2 YX2X20 X2X2PX2Y2,Y2_Y2_Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jJ  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1v bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1XQ J C V } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)I*I44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllncocqqsOtOvvx;y;{{}'~'uuaaבד b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2]Y2,]Y2/_Y2/_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 7  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 v a L 7 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2CW2TCW2͘W2͘W2W2W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2wX2 wX2X20 X2X2PX2cY2,cY2[`Y2[`Y2Y2$Y2Z2Z2/[2/[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S& bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S& bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S&  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 ;  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 < 5 .' } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)?*?44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33ww__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2Y2,Y2M_Y2M_Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 { b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations17w  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGItJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2GW2LGW2IW2TIW2ӘW2ӘW2 W2 W2{W2({W2W2W2W2W2X2(X2X2X2MX2MX2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2aX2PaX2Y2,Y2m`Y2m`Y2/Y2$/Y2Z2Z2A[2A[2eP\2eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_?   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 -M b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1K  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations104 8 <  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>3?3IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbdbebggiLjLlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2WY2,WY2`Y2`Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 l b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 -  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x n d Zn } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)@*@44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$l%l'')T*T,,.>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2{Y2,{Y23aY23aY2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S' bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S' bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S'  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 O  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@ D HL } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)>*>44>(?(IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI,J,LLNOQQSSUkVkXXZQ[Q]]_7`7bbdeggijlvlvnnp\q\ssuBvBxxz({(}}ggډڋMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2W2aW2$aW2KW2KW2W2W23W2$3W2X2X2[X2[X2`X2$`X27X27X2X2X2-X2 -X2UNY2| UNY2)Y2 )Y2Z2 Z2;[2 ;[2O\2 O\2!]2!]2q4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =k  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =k  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =k  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Gg  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  aa)K*K44>1?1IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&p p""$Z%Z'')D*D,,../.11346689;w;w==?]@]BBDCECGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~bbԉԋFF** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2uW2uW2W2W2W2W2GX2pGX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY25Y2 5Y2 Z2p  Z2G[2t G[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  ff)R*R44>:?:IIS"T"^^h i s|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvjj ^ ^  RRFF66""  "#%%'')p*p,,.\/\113H4H668494;;=>@@BCEzEzGGIbJbLLNJOJQQS2T2VVXY[[]^`v`vbbd^e^ggiFjFlln.o.qqstvvxxzr{r}}ZZ΂΄BB''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2W2aW2$aW2KW2KW2W2W23W2$3W2X2X2[X2[X2`X2$`X2X2X2X2X2-X2 -X2UNY2| UNY2Y2 Y2Z2 Z2#[2 #[2O\2 O\2c!]2c!]2 4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  __)E*E44>'?'IIS T ^z^zhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&h h""$N%N'')4*4,,./11346s6s88:Y;Y==?;@;BBDEGGIIKpLpNNPRQRSSU4V4XXZ[]]__aibiddfKgKiik-l-nnpqssuuwbxbzz|D}D&&vvVVƐƒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S5 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)J*J44>.?.IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvff V V  FF66""  ~ ~""$f%f'')N*N,,.6/611346689;z;z==?^@^BBDBEBGGI&J&LLN O Q|Q|SSU`V`XXZD[D]]_(`(bbd e g~g~iikblbnnpFqFssu*v*xxz{}}ddքֆFF((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2aW2aW2EW2EW2W2W23W23W2X2pX2[X2[X2`X2`X2MX2`MX2X2X2-X2-X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1^~  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  bb)L*L44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55""   ""$k%k'')U*U,,.?/?113)4)6689;;==?r@rBBD\E\GGIEJELLN-O-QQSTVVXXZq[q]]_Y`YbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%%mm b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2}X2`}X2X2X2-X2-X2ONY2L ONY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\< b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  dd)V*V44>F?FIIS2T2^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrdd V V  HH::,,  "#%%'(*{*{,,.m/m113_4_668Q9Q;;=C>C@@B5C5EEG'H'JJLMOOQRTzTzVVXhYh[[]V^V``bBcBeeg,h,jjlmooqqsotovvxUyU{{};~;!!ww[[͓͑ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2aW2aW2EW2EW2W2W23W23W2X2pX2[X2[X2`X2`X2WX2`WX2X2X2-X2-X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1l  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ^^)H*H44>0?0IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##m m""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI0J0LLNOQQSSUoVoXXZU[U]]_;`;bbdeggijlulunnpYqYssu=v=xxz!{!}}ww[[Ή΋AA'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2uX2`uX2X2X2-X2-X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\>  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\>  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\>  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ``)N*N44>:?:IIS"T"^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr`` N N  <<** } }""$k%k'')Y*Y,,.G/G113545668#9#;;=>@@BBDvEvGGIbJbLLNLOLQQS6T6VVX Y [[] ^ ``bbdeeeggiKjKlln/o/qqstvvxxzd{d}}FF(( zz\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2gW2gW2?W2?W2W2W29W29W2X2pX2[X2[X2 `X2 `X2?X2`?X2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>;?;IIS)T)^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIcJcLLNQOQQQS?T?VVX-Y-[[]^``b c eegginjnllnWoWqqs;t;vvxy{{}~uuYYˇˉ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2ݎW2ݎW2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2]X2`]X2X2X2'X2'X2ONY2L ONY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eE b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ^^)N*N44>>..  ""$v%v'')f*f,,.V/V113F4F668696;;=&>&@@BCEEGHJ|J|LLNhOhQQSTTTVVX@Y@[[],^,``bceeghjzjzllnfofqqsRtRvvx>y>{{}*~*xxddڑړ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2gW2gW29W29W2W2W29W29W2 X2p X2[X2[X2 `X2 `X2WX2`WX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  YY)C*C44>-?-IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ooYY C C  --vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLNOQQSSUnVnXXZX[X]]_B`Bbbd,e,ggijllnnpoqossuSvSxxz7{7}}qqUUǎǐ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2ݎW2ݎW2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2/X2`/X2X2X2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eG b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ee  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ZZ)F*F44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  l l""$X%X'')D*D,,.0/011346689;~;~==?j@jBBDVEVGGI@J@LLN(O(QQSTVVXXZl[l]]_T`Tbbd;?;IIS#T#^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqee Y Y  MMAA55))  "#%%' ( **,,.k/k113W4W668C9C;;=/>/@@BCEEGHJ{J{LLNcOcQQSKTKVVX3Y3[[]^``bcewewggi_j_llnGoGqqs/t/vvxy{{}}ss[[χωCC++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nN b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nN b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nN b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  XX)F*F44>2?2IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjXX F F  44""u u""$c%c'')Q*Q,,.?/?113-4-6689;;= > @@BBDnEnGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKlln5o5qqstvvx y {~{~}}hh݂݄RRLJlj<<&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2UW2UW23W23W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  __)G*G44>-?-IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??//  ""$s%s'')[*[,,.C/C113+4+6689;;==?o@oBBDWEWGGI=J=LLN!O!QQSTVwVwXXZ[[[]]_?`?bbd#e#ggijlylynnp]q]ssuAvAxxz%{%}} {{__щыCC'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nP b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nP b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nP b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  WW)A*A44>)?)IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //}}i i""$T%T'')>*>,,.(/(11346688:q;q==?[@[BBDEEEGGI-J-LLNOQQSSUlVlXXZR[R]]_8`8bbdeggijlwlwnnp]q]ssuCvCxxz){)}}hhۉۋNN44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2aW2aW23W23W2W2W23W23W2X2pX2[X2[X2`X2`X2'X2`'X2X2X2!X2!X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Dd  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  UU)?*?44>%?%IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?Q@QBBD7E7GGIJLLNOQvQvSSU\V\XXZB[B]]_(`(bbdeggiikglgnnpMqMssu3v3xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2׎W2׎W2W2W2+W2+W2W2W2_X2p_X2[X2[X2{bX2{bX2'X2`'X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wW# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wW# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wW# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)F*F44>.?.IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::**  "#%x%x'')d*d,,.P/P113<4<668(9(;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66tt b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2aW2aW2-W2-W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2 [2t  [2{O\2T {O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  SS)9*944>?IISS]n^nhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..vv\ \""$B%B'')(*(,,./11335g6g88:M;M==?/@/BBDEGGIIKdLdNNPFQFSSU(V(XXZ [ ]{]{__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2׎W2׎W2W2W2+W2+W2W2W2_X2p_X2[X2[X2{bX2{bX2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY% b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY% b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wY% b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)>*>44>"?"IIST^x^xhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**r r""$Z%Z'')B*B,,.*/*11346688:n;n==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2OW2OW2-W2-W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  VV)@*@44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99)) u u""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>:?:IIS&T&^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffXX J J  <<..    "#%}%}'')o*o,,.a/a113S4S668E9E;;=7>7@@B)C)EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m o}o}qqsctcvvxIyI{{}/~/kk݌ݎOO b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2OW2OW2'W2'W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2EX2EX2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 *  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>$?$IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::nnߎߐPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2[W2[W2!W2!W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2)[2t )[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2ˎW2ˎW2W2W21W21W2W2W2eX2peX2\X2\X2obX2obX2X2`X2uX2uX2X2X2=NY2L =NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)B*B44>0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$j%j'')Z*Z,,.J/J113:4:668*9*;;=>@@B C EEGGIpJpLLN\O\QQSHTHVVX4Y4[[] ^ ``b c eegginjnllndodqqsPtPvvx!?!IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2ˎW2ˎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2obX2obX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>$?$IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hbbd0e0ggijllnoq|q|ssudvdxxzL{L}}44xx`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2[X2[X2_X2_X2GX2`GX2X2X2X2X27NY2T 7NY2Y2 Y2EZ2x EZ2#[2| #[2SP\2\ SP\2I ]2I ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1E(  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  YY)G*G44>1?1IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUllnIoIqqs1t1vvxy{{}~uu]]чщEE-- b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2eW2 eW2ݎW2ݎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2SX2SX2X2X27NY2T 7NY2#Y2 #Y2KZ2x KZ2)[2| )[2YP\2\ YP\2O ]2O ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1?  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  LL)D*D44>0?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln?o?qqs)t)vvxy{{}}rr\\чщFF00 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2IGW2IGW2W2 W2IW2IW2W2W2W2W2W2W2X2pX2[X2[X2_X2_X2X2`X2cX2cX2X2X27NY2T 7NY2Y2 Y2-Z2x -Z2 [2|  [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1K  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)?*?44>%?%IIS T ^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpaqassuEvExxz){)}} ccՉՋGG++ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2kW2 kW2ݎW2ݎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2X2X27NY2T 7NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1/:  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  KK)=*=44>%?%IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##qq] ]""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpeqessuKvKxxz1{1}}ppVVɎɐ<< b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2 W2 W2?W2?W2əW2|əW2W2HW2ٰW2ٰW2wW2PwW2W2XW2W2W2X2PX2X2X2X2X2I[X2$ I[X2`X2, `X2cX2 cX2mmX2$ mmX2pX2 pX2X2| X2MX2XMX2X2X2X2X2qY2qY2-=Y2\-=Y2]Y2]Y2hY2hY2Y28Y2Z2Z2/[2/[2[2[2SP\2lSP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1U  [  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dU *yRT+)~S(}R' | Q &  {  P  :zd bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  )D*D44>2?2IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  ((CC1 1""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKllnAoAqqs+t+vvxy{{}}tt^^ӇӉHH22# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2}W2d}W21W2l1W2W2W2W2W2W2@W2OW2OW2W2W2X2@X2X2 X2X2X2!\X2!\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2kX2 kX2X2H X2 X2h X2Y2DY2='?'IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$#%#'') * ,,..01346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;##mm  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2 W2 W2?W2?W2əW2|əW2W2HW2ӰW2ӰW2qW2PqW2W2XW2W2W2X2PX2X2X2X2X2C[X2$ C[X2`X2, `X2cX2 cX2mmX2$ mmX2pX2 pX2X2| X2SX2XSX2X2X2X2X2)Y2)Y2'?'IIS T ^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##88$ $""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnpgqgssuMvMxxz3{3}}rrXXˎː>># b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW21W2l1W2W2W2W2W2W2@W2UW2UW2W2W2X2@X2X2 X2X2X2'\X2'\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2ץX2 ץX27X2H 7X2yX2hyX2Y2DY2s$?$IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$%''))+o,o..01346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggijlulunnpYqYssu=v=xxz!{!}}ww]]ЉЋCC))  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2IGW2IGW2W2dW21W2l1W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2X2 X2X2H X2%X2h%X2Y2DY2y*>44>(?(IdIdST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22""  ""$p%p'')\*\,,.H/H113444668 9 ;;= > @@BBDhEhGGIPJPLLN8O8QQS T V\V\XXZD[D]]_,`,bbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2wW2dwW21W2l1W2W2W2W2W2W2@W2IW2IW2W2W2X2@X2X2 X2}X2}X2\X2\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2kX2 kX2X2H X2 X2h X2uY2DuY21/?/IIST^Y^Yhisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnoqqssukvkxxz{}}kk݂݄OO33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2IGW2IGW2W2dW21W2l1W2W2W2W2W2W2@W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2X2 X2[X2H [X2X2hX2Y2DY2U > ~  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dbD&jjLE. pP0   p  P  0  rpTT bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)6*644>?IUIUST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""j j""$R%R''):*:,,."/"113 4 6~6~88:f;f==?L@LBBD0E0GGIJLLNNPjQjSSUVXXZZ\k]k__abdegtgtiikXlXnnp>   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2}W2d}W21W2l1W2W2W2W2W2W2@W2IW2IW2W2W2X2@X2X2 X2}X2}X2\X2\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX21X2 1X2X2H X2X2hX2]Y2D]Y2^^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1U*/oo  D  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dU*T) ~jSG($}R' | Q &  {  P  :zd bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^H^Hhhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnp.q.ssuvxxz/{/}}iiۉۋMM11  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW27W2T7W2W2W2W2W2!W2(!W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2yX2 yX2X20 X2X2PX2_X2,_X2k\Y2k\Y2Y2$Y2Z2Z2/[2/[2wP\2wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TU* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TU* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TU*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s6 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1yV3 , } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  TT)F*F44>8?8IIS$T$^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbTT F F  88**  "#%y%y'')k*k,,.]/]113O4O668A9A;;=3>3@@B%C%EEGHJJLMO~O~QQSlTlVVXZYZ[[]H^H``b4c4eeghjjlmo{o{qqsatavvxGyG{{}-~-kk݌ݎOO b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2=W2T=W2ǘW2ǘW2W2W2W2(W2CW2CW2W2W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X2ѥX2 ѥX21X20 1X2sX2PsX2gX2,gX2;\Y2;\Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 L bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 gB  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2OGW2OGW2W2LW27W2T7W2W2W2W2W2W2(W2W2W2 W2 W2X2(X2X2X2X2X2[X2[X2bX2 bX2elX2 elX2oX2L oX2ݜX2 ݜX2X2 X2 X20  X2OX2POX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UU* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UU* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UU*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2R b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mJ' } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbdWeWggi=j=lln!o!qqstvtvtxxzV{V}}88nnߎߐPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2=W2T=W2ǘW2ǘW2W2W2W2(W2CW2CW2W2W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2aX20 aX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ++K b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xS. W } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2X20 X21X2P1X2sX2,sX2\Y2\Y2#Y2$#Y2Z2Z25[25[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @W, bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @W, bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @W,  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x @ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rQ0C } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR)B*B44>2?2IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$j%j'')Z*Z,,.J/J113:4:668*9*;;=>@@B C EEGGIrJrLLN^O^QQSJTJVVX6Y6[[]"^"``bceeggipjplln\o\qqsHtHvvx4y4{{} ~  nnZZБГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2IW2TIW2ӘW2ӘW2 W2 W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2[X2,[X2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U9 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U9 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o;  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U9  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 <Aa b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Z4  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2!W2(!W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2UX20 UX2X2PX2 Y2, Y2%]Y2%]Y2 Y2$ Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ? b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1nM, } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN):*:44>&?&IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqvqvssu^v^xxzF{F}}..rrZZ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2=W2=W2W2W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2ۤX2 ۤX2;X20 ;X2}X2P}X2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s?  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ( b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jDx4t } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2UW2TUW2ߘW2ߘW2W2W2'W2('W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2QX2PQX2MX2,MX2\Y2\Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  ImB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  ImB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  ImB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }*J b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1) bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 lM.[ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2CW2TCW2͘W2͘W2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2äX2 äX2#X20 #X2eX2PeX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U- bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U- bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U-  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1eJ/@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  TT)>*>44>)?)IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''s s""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2UW2TUW2ߘW2ߘW2W2W2'W2('W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2Y2,Y2C]Y2C]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JlA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JlA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JlA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 c b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pQ2 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2CW2TCW2͘W2͘W2W2W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2Y2,Y2m]Y2m]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Lll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1y^C% } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2 X2  X2kX20 kX2X2PX2GX2,GX2\Y2\Y2Y2$Y2Z2Z2)[2)[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ',L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1" bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1eF'T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLN3?3IIS!T!^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGI[J[LLNIOIQQS7T7VVX%Y%[[]^``bcexexggifjfllnOoOqqs3t3vvxy{{}}mm߂߄QQÇÉ55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2/X2 /X2X20 X2X2PX2Y2,Y2C]Y2C]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,LQq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1iJ+ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2CW2TCW2͘W2͘W2W2W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X27X2P7X2Y2,Y2-^Y2-^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y])  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 (H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2qX20 qX2X2PX2gX2,gX21]Y21]Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 (y-M b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1iR; } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2%W2%W2W2W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2gX2,gX2-^Y2-^Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1lN  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1{vq } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2YX2 YX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  y b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$ v_I } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2%Y2,%Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U* bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U* bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U*  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  n b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lgb][ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>)?)IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpYqYssu=v=xxz!{!}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W29W2(9W2W2W2W2W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2qX2 qX2X20 X2X2PX2;X2,;X2]Y2]Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1th\P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2MW2LMW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2CX2PCX2mX2,mX2^Y2^Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aC  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 g b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 P } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?Q@QBBD7E7GGIJLLNOQvQvSSU\V\XXZB[B]]_(`(bbdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W29W2(9W2W2W2W2W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2/[2/[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  n b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PD8,D } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2W2W2W2W2X2(X2X2X2SX2SX2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2+X2 +X2X20 X2X2PX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2[2[2O\2O\2W!]2W!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U+ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U+ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U+  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :Z>>^ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$*V } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  QQ)7*744>?IISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?/@/BBDEGGIIKdLdNNPFQFSSU(V(XXZ [ ]{]{__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2#W2aW2$aW2QW2QW2W2W23W2$3W2#X2#X2[X2[X2`X2$`X2X2X2X2X2-X2 -X2UNY2| UNY2AY2 AY2Z2 Z2S[2 S[2wP\2 wP\2G"]2G"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations19m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdnenggiXjXllnBoBqqs,t,vvxy{{}~uu__ԇԉII33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2GW2 GW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U7 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  aa)K*K44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44!!   ""$j%j'')T*T,,.>/>113(4(6689;;==?q@qBBD[E[GGIDJDLLN,O,QQSTVVXXZp[p]]_X`Xbbd@e@ggi(j(llnoqqssulvlxxzT{T}}<<$$ ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2W2aW2$aW2QW2QW2W2W23W2$3W2#X2#X2[X2[X2`X2$`X2X2X2oX2oX2-X2 -X2UNY2| UNY2/Y2 /Y2Z2 Z2A[2 A[2P\2 P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:k  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:k  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1:k  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)L*L44>4?4IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI8J8LLNOQQSTVwVwXXZ][]]]_C`Cbbd)e)ggijllnnphqhssuNvNxxz4{4}}ssYY̎̐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2W2W2{W2{W2W2W2W2W2MX2pMX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1U9 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  __)I*I44>1?1IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI1J1LLNOQQSSUpVpXXZV[V]]_<`v>xxz"{"}}xx\\ωϋBB(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2aW2aW2KW2KW2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2-X2-X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  gg)S*S44>;?;IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG77##  "#%%'')q*q,,.]/]113I4I668595;;=>@@BCE{E{GGIcJcLLNKOKQQS3T3VVXY[[]^`w`wbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2MW2 MW2W2W2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2X2`X2X2X2-X2-X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^@  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^@  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>:?:IIS(T(^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGIbJbLLNPOPQQS>T>VVX,Y,[[]^``bceeggimjmllnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2aW2aW2EW2EW2W2W23W23W2X2pX2[X2[X2`X2`X2]X2`]X2X2X2-X2-X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  aa)I*I44>-?-IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55!!  } }""$e%e'')M*M,,.5/511346689;y;y==?]@]BBDAEAGGI%J%LLN O Q{Q{SSU_V_XXZC[C]]_'`'bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2uX2`uX2X2X2-X2-X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^B b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^B b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^B b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ZZ)D*D44>.?.IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppZZ D D  ..wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBDG?GIIS3T3^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssee W W  II;;--  "#%%'(*|*|,,.n/n113`4`668R9R;;=D>D@@B6C6EEG(H(JJLMOOQRT{T{VVXiYi[[]W^W``bCcCeeg-h-jjlmooqqsptpvvxVyV{{}<~<""xx\\ΑΓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2ݎW2ݎW2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2WX2`WX2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gI b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gI b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gI b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1n  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  dd)P*P44>:?:IIS"T"^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@44((  "#%%'(*~*~,,.j/j113V4V668B9B;;=.>.@@BCEEGHJzJzLLNbObQQSJTJVVX2Y2[[]^``bcevevggi^j^llnFoFqqs.t.vvxy{{}}rrZZ·ΉBB** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2gW2gW2?W2?W2W2W29W29W2X2pX2[X2[X2 `X2 `X2]X2`]X2X2X2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1r  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS!T!^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbddedggiJjJlln.o.qqstvvxxzc{c}}EE''yy[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2ݎW2ݎW2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gK b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)H*H44>.?.IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00    "#%t%t'')\*\,,.D/D113,4,6689;;==?p@pBBDXEXGGI>J>LLN"O"QQSTVxVxXXZ\[\]]_@`@bbd$e$ggijlzlznnp^q^ssuBvBxxz&{&}} ||``҉ҋDD(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2UW2UW2?W2?W2W2W2'W2'W2X2pX2[X2[X2_X2_X2?X2`?X2X2X2!X2!X2INY2L INY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  __)O*O44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??//  ""$w%w'')g*g,,.W/W113G4G668797;;='>'@@BCEEGHJ}J}LLNiOiQQSUTUVVXAYA[[]-^-``bceeghj{j{llngogqqsStSvvx?y?{{}+~+yyeeۑۓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2!X2`!X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pR b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pR b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pR b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  TT)>*>44>$?$IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiUU A A  --yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2UW2UW29W29W2W2W2'W2'W2 X2p X2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)E*E44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$W%W'')C*C,,.///11346689;};}==?i@iBBDUEUGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77{{cc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2 X2 X2!X2!X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pT  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pT  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pT  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  TT):*:44>?IISS]o^ohhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //ww] ]""$C%C''))*),,./11335h6h88:N;N==?0@0BBDEGGIIKeLeNNPGQGSSU)V)XXZ [ ]|]|__a^b^ddf@g@iik"l"nnpqsusuuuwWxWzz|9}9kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2gW2gW29W29W2W2W29W29W2 X2p X2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbdbebggiLjLlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2׎W2׎W2W2W2+W2+W2W2W2_X2p_X2[X2[X2{bX2{bX2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y[' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2aW2aW23W23W2W2W23W23W2X2pX2[X2[X2`X2`X2}X2`}X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)@*@44>(?(IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$S%S'')=*=,,.'/'11346688:p;p==?Z@ZBBDDEDGGI,J,LLNOQQSSUkVkXXZQ[Q]]_7`7bbdeggijlvlvnnp\q\ssuBvBxxz({(}}ggډڋMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2׎W2׎W2W2W21W21W2W2W2eX2peX2\X2\X2{bX2{bX2X2`X2X2X2!X2!X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>%?%IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2OW2OW2-W2-W2W2W2!W2!W2X2pX2[X2[X2_X2_X2-X2`-X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ll  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>/?/IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbdSeSggi;j;lln#o#qqs t vvxxzg{g}}OOÂÄ77uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2OW2OW2-W2-W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  UU)=*=44>!?!IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2٣X2`٣X29X29X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f2 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2#Y2 #Y2Z2p Z25[2t 5[2}P\2T }P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>;?;IIS'T'^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2ˎW2ˎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2obX2obX2X2`X2oX2oX2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16V  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  XX)D*D44>.?.IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44((  "#%%'')r*r,,.^/^113J4J668696;;=">"@@BCEEGGInJnLLNVOVQQS>T>VVX&Y&[[]^``bbdjejggiRjRllnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2UGW2UGW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2ߣX2`ߣX2?X2?X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1&  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2ˎW2ˎW2W2W27W27W2W2W2kX2pkX2 \X2 \X2obX2obX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o; b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)<*<44>"?"IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI2J2LLNOQQSSUlVlXXZP[P]]_4`4bbdeggiiknlnnnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2OGW2OGW2W2 W2IW2IW2!W2!W2W2W2W2W2X2pX2[X2[X2_X2_X2MX2`MX2X2X2X2X27NY2T 7NY2)Y2 )Y2QZ2x QZ2/[2| /[2P\2\ P\2 ]2 ]2-3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1G,  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)E*E44>3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqllnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2kW2 kW2ݎW2ݎW2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2 Y2  Y23Z2p 3Z2[2t [2O\2T O\2]2]2y2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10]]  RR)<*<44>"?"IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1]]II 5 5  !!  wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbdeggiikplpnnpVqVssu'?'IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2qW2 qW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2wX2`wX2X2X2X2X27NY2L 7NY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2k]2k]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  PP)6*644>?IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2 GW2 GW2W2|W2W2W2EW2EW2ϙW2|ϙW2W2HW2ٰW2ٰW2wW2PwW2W2XW2W2W2X2PX2X2X2X2X2I[X2$ I[X2`X2, `X2cX2 cX2smX2$ smX2pX2 pX2X2| X25X2X5X2X2X2X2X2 Y2 Y2Y>Y2\Y>Y2 ^Y2 ^Y2iY2iY2Y28Y2Z2Z2/[2/[2[2[2SP\2lSP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W(h  D D  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dW.\3 a8ff?=    q k B   p  ^5  bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  )K*K44>7?7IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11eeY Y""$%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnOoOqqs7t7vvxy{{}~{{ccׇ׉KK33# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW21W2l1W2W2W2W2W2W2@W2UW2UW2W2W2X2@X2X2 X2X2X2'\X2'\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2ӦX2 ӦX23X2H 3X2uX2huX2g Y2Dg Y2#>Y2#>Y2 ^Y2 ^Y2iY2 iY2#Y2#Y2Z2pZ25[2t5[2[2T[2}P\2}P\2M"]2$M"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Wq. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Wq. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Wq.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb" b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W.6v v  M b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dW.\3 a8@fj= ?   k B   p  ^5 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsqtqvvxWyW{{}=~=## {{__ёѓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2W2W2EW2EW2ϙW2|ϙW2W2HW2ٰW2ٰW2wW2PwW2W2XW2W2W2X2PX2X2X2X2X2I[X2$ I[X2`X2, `X2cX2 cX2smX2$ smX2pX2 pX2X2| X2SX2XSX2X2X2X2X2E Y2E Y2?Y2\?Y2 ^Y2 ^Y2iY2iY2Y28Y2Z2Z2[2[2[2[2O\2lO\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1(X  4 b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dW.\3 a8f=   k J B #  |  p U  ^5y bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  )A*A44>)?)IIS T ^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!NN> >""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI9J9LLNOQQSTVsVsXXZW[W]]_;`;bbdeggijlulunnpeqessuIvIxxz-{-}}ggىًKK//# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW21W2l1W2W2W2W2W2W2@W2UW2UW2W2W2X2@X2X2 X2X2X2'\X2'\X2bX2 bX2_lX2 _lX2oX2d oX2לX2 לX2/X2 /X2X2H X2X2hX2Y2DY2  ]   @J'g- bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)B*B44>0?0IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$$%$'')*,,./113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ~~``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2OGW2OGW2W2dW27W2l7W2W2W2W2W2!W2@!W2W2W2 W2 W2X2@X2X2 X2X2X2[X2[X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2X2 X2X2H X2+X2h+X2Y2DY2y?ITITS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^JJ 6 6  ""nnX X""$B%B''),*,,,./11346u6u88:_;_==?G@GBBD-E-GGIJLLNNPlQlSSUVXXZZ\q]q__abdeggiikglgnnpMqMssu3v3xxz{}}rrXXʉʋ<<   b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2}W2d}W27W2l7W2W2W2W2W2W2@W2OW2OW2 W2 W2X2@X2X2 X2X2X2!\X2!\X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2X2 X2X2H X2EX2hEX2iY2DiY2%4?4IIS T ^^^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln&o&qqstvvxxz{}$~$rr^^ԑԓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2OGW2OGW2W2dW27W2l7W2W2W2W2W2W2@W2W2W2 W2 W2X2@X2X2 X2X2X2[X2[X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2X2 X2aX2H aX2X2hX2 Y2D Y2G>Y2G>Y2_Y2_Y2jY2 jY2Y2Y2Z2pZ2[2t[2y[2Ty[2oO\2oO\2?!]2$?!]23]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s60P b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1mZ 2 2 r b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dmZG4!{hS>  )   g  B  j x U @-m0p bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  II)/*/44>?IIIISS]n^nhhrPsP} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$llR R""$8%8'')*,,./1w1w335]6]88:C;C==?'@'BBD E GzGzIIK\L\NNP>Q>SSUUWVXVZZ\8]8__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8ll܋܍LL  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW27W2l7W2W2W2W2W2W2@W2OW2OW2 W2 W2X2@X2X2 X2X2X2!\X2!\X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2SX2 SX2X2H X2X2hX2Y2DY2y&?&IIST^K^Khhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$L%L'')8*8,,.$/$11346688:r;r==?^@^BBDJEJGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnnp?q?ssu'v'xxzJ{J}}22vv^^  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2+X2 +X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UW. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UW. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UW.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 oQq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1<|  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1_<  $d } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)8*844>#?#IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!m m""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI3J3LLNOQQSTVwVwXXZ_[_]]_G`Gbbd/e/ggijllnnpsqsssu[v[xxzC{C}}++ss]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2=W2T=W2ǘW2ǘW2W2W2W2(W2CW2CW2W2W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2klX2 klX2oX2L oX2X2 X2 X2  X2mX20 mX2X2PX2Y2,Y2I]Y2I]Y2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 66V b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1$d  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1N' Y  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2=W2T=W2ǘW2ǘW2W2W2!W2(!W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2klX2 klX2oX2L oX2X2 X2X2 X2 X20  X2OX2POX23Y2,3Y2?^Y2?^Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 VW. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 VW. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 VW.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x @ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1v S 0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL)6*644> ? IIST^w^whhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikelennpIqIssu-v-xxz{}}ggلنMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2CW2TCW2͘W2͘W2W2W2W2(W2IW2IW2W2W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2CX2 CX2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w )n b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations12 d = a } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)9*944>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2/X2 /X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Y0 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Y0 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Y0  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 pTt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1O bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 hG & ? } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIUJULLNCOCQQS1T1VVXY[[] ^ ``bbdrerggi`j`llnIoIqqs-t-vvxy{{}}ggقلKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2OW2TOW2٘W2٘W2W2W2W2(W2=W2=W2!W2!W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W= bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W= bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s?  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W=  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ccc b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^<  0p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2CW2TCW2͘W2͘W2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2ѥX2 ѥX21X20 1X2sX2PsX23Y2,3Y2c^Y2c^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }*J b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  d C "  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>!?!IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQxQxSSUbVbXXZL[L]]_6`6bbd e ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2OW2TOW2٘W2٘W2W2W2W2(W2CW2CW2!W2!W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2AX2 AX2X20 X2X2PX2;Y2,;Y2Q^Y2Q^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 rHhm b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x V 4   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44> ? IIST^v^vhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?P@PBBD4E4GGIJLLNNPnQnSSURVRXXZ6[6]]_`bbddfpgpiikTlTnnp8q8ssuvxxz{}r}rVVȄȆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2[W2T[W2W2W2W2W2-W2(-W2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2_X20 _X2X2PX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IoF bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IoF bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IoF  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q!!A b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1)i  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rS 4 [ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  YY)E*E44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%%'')s*s,,._/_113K4K668797;;=#>#@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2IW2TIW2ӘW2ӘW2 W2 W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2+X2P+X2Y2,Y2^Y2^Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W1 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W1  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 < b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1XA*  6 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  ZZ)L*L44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhZZ L L  >>00""  "#%%'')q*q,,.c/c113U4U668G9G;;=9>9@@B+C+EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3qqUUǑǓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2[W2T[W2W2W2W2W2-W2(-W2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2-Y2,-Y2^Y2^Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JnE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JnE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JnE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 h b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  v W 8 < } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)=*=44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$i%i'')Q*Q,,.9/9113!4!668 9 ;};}==?e@eBBDMEMGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2=W2=W2W2W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2)X2P)X29Y2,9Y2)_Y2)_Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 | e N 7 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<rrTT b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 2h7W b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"b  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 kL - T } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbdeg{g{iikalannpGqGssu-v-xxz{}}ll߄߆RRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2IW2TIW2ӘW2ӘW2 W2 W2W2(W21W21W2W2W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2IX2PIX2Y2,Y2G_Y2G_Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1y])  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 @` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1l bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10$   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>8?8IIS$T$^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJxJxLLNdOdQQSPTPVVX?IISS]j^jhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?+@+BBD E G~G~IIK`L`NNPBQBSSU$V$XXZ[]w]w__aYbYddf;g;iiklnnpprpspuuwRxRzz|4}4hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2IW2TIW2ӘW2ӘW2 W2 W2W2(W27W27W2W2W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2ѥX2 ѥX21X20 1X2sX2PsX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a-  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 `  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 t h \p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)>*>44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$P%P'')<*<,,.(/(11346689;v;v==?b@bBBDNENGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2/[2/[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  n b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1,l  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1H3 2 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2Y2,Y2`Y2`Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1nR  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  g b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1J bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbdaeaggiKjKlln5o5qqstvvx y {~{~}}hh݂݄RRLJlj<<&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2qX2 qX2X20 X2X2PX2?Y2,?Y2/_Y2/_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 t _ J 5 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>$?$IIS T ^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bbdeggiikilinnpMqMssu1v1xxz{}}kk݄݆QQĉċ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2IW2TIW2ӘW2ӘW2 W2 W2W2(W2+W2+W2W2W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2#X2 #X2X20 X2X2PX2sY2,sY2`Y2`Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W. bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W. bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W.  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9Kk>^ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 I  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  A } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>'?'IIS T ^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggijlulunnp[q[ssuAvAxxz'{'}} ffىًLL22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2OW2TOW2٘W2٘W2W2W2?W2(?W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2Y2,Y2M_Y2M_Y2)Y2$)Y2Z2Z2;[2;[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  g b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations14t  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1    } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2SW2LSW2OW2TOW2٘W2٘W2W2W2W2(W2%W2%W2!W2!W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2X20 X27X2P7X2sY2,sY2`Y2`Y2)Y2$)Y2Z2Z2;[2;[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1cG  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 E  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PX ` h $d } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<,,  "#%z%z'')f*f,,.R/R113>4>668*9*;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2WY2,WY2`Y2`Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 *  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x n d Zn } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  OO)9*944>%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeOO 9 9  ##  llV V""$@%@'')***,,./11335s6s88:];]==?G@GBBD1E1GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2OW2TOW2٘W2٘W2W2W2W2(W2%W2%W2!W2!W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2ťX2 ťX2%X20 %X2gX2PgX2= Y2,= Y2aY2aY2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W/ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W/ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W/  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 \| b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p x , } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)>*>44>$?$IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**r r""$Z%Z'')B*B,,.*/*11346688:n;n==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp>  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2#W2gW2$gW2WW2WW2W2W29W2$9W2)X2)X2[X2[X2 `X2$ `X2X2X2X2X23X2 3X2UNY2| UNY2AY2 AY2Z2 Z2S[2 S[2wP\2 wP\2G"]2G"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations15o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations15o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations15o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  nn)Z*Z44>D?DIIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ>>22  "&#&%%'(**,,.t/t113`4`668L9L;;=8>8@@B$C$EEGHJJLLNlOlQQSTTTVVXI?IIIS5T5^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uugg Y Y  KK==//!!  "#%%'(*~*~,,.p/p113b4b668T9T;;=F>F@@B8C8EEG*H*JJLMOOQRT}T}VVXkYk[[]Y^Y``bEcEeeg/h/jjlmooqqsrtrvvxXyX{{}>~>$$zz^^БГ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2#W2gW2$gW2QW2QW2W2W29W2$9W2#X2#X2[X2[X2 `X2$ `X2X2X2X2X23X2 3X2UNY2| UNY2)Y2 )Y2Z2 Z2;[2 ;[2O\2 O\2!]2!]2w4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations16m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)P*P44>6?6IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%|%|'')d*d,,.L/L1134446689;;=>@x@xBBD`E`GGIFJFLLN*O*QQSTVVXXZd[d]]_H`Hbbd,e,ggijllnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2W2W2W2W2SX2pSX2[X2[X2bX2bX2uX2`uX2X2X23X23X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W=  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W=  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1W=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  cc)Q*Q44>=?=IIS%T%^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uucc Q Q  ??--   ""$n%n'')\*\,,.J/J113848668&9&;;=>@@BCEyEyGGIeJeLLNOOOQQS9T9VVX#Y#[[] ^ ``bbdhehggiNjNlln2o2qqstvvxxzg{g}}II++ }}__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2aW2aW2QW2QW2W2W23W23W2#X2p#X2[X2[X2`X2`X2WX2`WX2X2X2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1i  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ^^)H*H44>.?.IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##m m""$W%W'')A*A,,.+/+11346688:t;t==?Z@ZBBD@E@GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpVqVssu???IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11!!  "#%y%y'')i*i,,.Y/Y113I4I668999;;=)>)@@BCEEG H JJLLNkOkQQSWTWVVXCYC[[]/^/``bceeghj}j}llnioiqqsUtUvvxAyA{{}-~-{{ggݑݓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2aW2aW2KW2KW2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2c!]2c!]2 4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  \\)B*B44>$?$IIST^w^whhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##  e e""$K%K'')1*1,,./11335p6p88:V;V==?8@8BBDEGGIIKmLmNNPOQOSSU1V1XXZ[]]__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##ssSSÐÒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`F b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`F b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`F b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])I*I44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$[%['')G*G,,.3/31134668 9 ;;==?m@mBBDYEYGGICJCLLN+O+QQSTVVXXZo[o]]_W`Wbbd?e?ggi'j'llnoqqssukvkxxzS{S}};;## gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2gW2gW2KW2KW2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  __)I*I44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22  ~ ~""$h%h'')R*R,,.e>ggi&j&llnoqqssujvjxxzR{R}}::"" jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2ݎW2ݎW2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2WX2`WX2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iM b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iM b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iM b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>5?5IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI]J]LLNGOGQQS1T1VVXY[[]^`z`zbbddedggiNjNlln8o8qqs"t"vvx y {{}}kkUUʇʉ??)) b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2gW2gW2EW2EW2W2W29W29W2X2pX2[X2[X2 `X2 `X2cX2`cX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1u  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>-?-IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?[@[BBDEEEGGI-J-LLNOQQSSUlVlXXZR[R]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXˉˋ>>$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2ݎW2ݎW2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2eX2`eX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iO b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iO b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iO b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1w  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ZZ)D*D44>,?,IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  l l""$W%W'')A*A,,.+/+11346688:t;t==?^@^BBDHEHGGI0J0LLNOQQSSUoVoXXZU[U]]_;`;bbd!e!ggijlzlznnp`q`ssuFvFxxz,{,}}kkމދQQĎĐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2[W2[W2EW2EW2W2W2-W2-W2X2pX2[X2[X2_X2_X2EX2`EX2X2X2'X2'X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1_  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>8?8IIS&T&^^h i s|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI`J`LLNNONQQS1?1IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==--  "#%{%{'')g*g,,.S/S113?4?668+9+;;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99ww b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  VV)@*@44>*?*IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llVV @ @  **ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLN O QQSSUkVkXXZU[U]]_?`?bbd)e)ggijllnnplqlssuPvPxxz4{4}}nnRRĎĐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2 X2 X2'X2'X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rX$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rX$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rX$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)A*A44>%?%IIS T ^{^{hhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==-- u u""$]%]'')E*E,,.-/-11346688:q;q==?U@UBBD9E9GGIJLLNOQsQsSSUWVWXXZ;[;]]_`bbdeguguiikYlYnnp=q=ssu!v!xxz{}w}w[[̈́͆== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2gW2gW29W29W2W2W29W29W2 X2p X2[X2[X2 `X2 `X2?X2`?X2X2X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Zz  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>22&&  "#%%'(*|*|,,.h/h113T4T668@9@;;=,>,@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2׎W2׎W2W2W21W21W2W2W2eX2peX2\X2\X2{bX2{bX2X2`X2 X2 X2!X2!X2INY2L INY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{_+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  [[)M*M44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii[[ M M  ??11##  "#%%'')r*r,,.d/d113V4V668H9H;;=:>:@@B,C,EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2gW2gW29W29W2W2W29W29W2 X2p X2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)D*D44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$p%p'')X*X,,.@/@113(4(6689;;==?l@lBBDTETGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2׎W2׎W2W2W27W27W2W2W2kX2pkX2 \X2 \X2{bX2{bX2ߣX2`ߣX2?X2?X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2OW2OW23W23W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>"?"IIST^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2OW2OW23W23W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]k^khhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>&?&IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00vv`` b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2ˎW2ˎW2W2W2=W2=W2W2W2qX2pqX2\X2\X2obX2obX2X2`X2uX2uX2X2X2=NY2L =NY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>)?)IIST^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGIQJQLLN;O;QQS%T%VVXY[[]]_n`nbbdXeXggiBjBlln6o6qqs t vvx y {{}}iiނބSSȇȉ=='' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2[W2[W2'W2'W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2KX2KX2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikjljnnpNqNssu2v2xxz{}}llބކPPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2ˎW2ˎW2W2W2=W2=W2W2W2qX2pqX2\X2\X2obX2obX2X2`X2X2X2X2X2=NY2L =NY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844> ? IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnVoVqqs:t:vvxy{{}~ttXXʇʉ<<  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2}X2`}X2X2X2X2X2=NY2L =NY2Y2 Y2EZ2p EZ2#[2t #[25P\2T 5P\2+ ]2+ ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)G*G44>/?/IIS#T#^^h i s}s}} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11!!   ""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDoEoGGIWJWLLN?O?QQS'T'VVXY[[]]_k`kbbd_e_ggiGjGlln/o/qqstvvxxzs{s}}[[ςτCC(( b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2UGW2UGW2W2 W2IW2IW2!W2!W2W2W2W2W2X2pX2[X2[X2_X2_X2)X2`)X2X2X2X2X2=NY2T =NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1p  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  JJ)8*844>"?"IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnppqpssuTvTxxz8{8}}rrVVȎȐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2ͣX2`ͣX2-X2-X2X2X2=NY2L =NY2Y2 Y2-Z2p -Z2 [2t  [2O\2T O\2]2]2[2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)=*=44>!?!IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!  i i""$Y%Y'')A*A,,.)/)11346688:m;m==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbd e g}g}iikalannpEqEssu)v)xxz { }}ccՄՆEE''  b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2 GW2 GW2W2|W2W2W2EW2EW2ϙW2|ϙW2W2HW2߰W2߰W2}W2P}W2W2XW2W2W2X2PX2X2X2X2X2O[X2$ O[X2`X2, `X2cX2 cX2smX2$ smX2pX2 pX2X2| X2SX2XSX2X2X2X2X2{Y2{Y279?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //KK; ;""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllnoooqqs[t[vvxGyG{{}3~3 mm# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW27W2l7W2W2W2W2W2W2@W2[W2[W2 W2 W2X2@X2X2 X2X2X2-\X2-\X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2X2 X2X2H X2EX2hEX2{Y2D{Y270?0IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$!%!'')*,,..013'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnXoXqqs>""  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2 GW2 GW2W2|W2W2W2EW2EW2ϙW2|ϙW2W2HW2ٰW2ٰW2wW2PwW2W2XW2W2W2X2PX2X2X2X2X2I[X2$ I[X2`X2, `X2cX2 cX2smX2$ smX2pX2 pX2X2| X2;X2X;X2X2X2X2X2iY2iY2%+?+IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##44 ""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoqqssuovoxxzW{W}}??''kk# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW27W2l7W2W2W2W2W2W2@W2[W2[W2 W2 W2X2@X2X2 X2X2X2-\X2-\X2bX2 bX2elX2 elX2oX2d oX2ݜX2 ݜX2X2 X2OX2H OX2X2hX2uY2DuY21$?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``JJ 4 4  }}ggQ Q""$%'w'w))+a,a..01335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2UGW2UGW2W2dW2=W2l=W2ǘW2ǘW2W2W2'W2@'W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2klX2 klX2oX2d oX2X2 X2YX2 YX2X2H X2X2hX2 Y2D Y2I?Y2I?Y29^Y29^Y27iY2 7iY2Y2Y2Z2pZ2)[2t)[2[2T[2MP\2MP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Yl?  / o b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dY2r Kd$=oH!zS  ,   ^ 7   ) [ bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  MM);*;44>)?)IISS]^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGIQJQLLN;O;QQS%T%VVXY[[]]_n`nbbdeggijl{l{n6o6qqs t vvx y {{}}iiނބSSȇȉ==''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2=W2l=W2ǘW2ǘW2W2W2W2@W2OW2OW2W2W2X2@X2X2 X2X2X2!\X2!\X2bX2 bX2klX2 klX2oX2d oX2X2 X2X2 X2X2H X2+X2h+X2#Y2D#Y22?2IIST^T^Thisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVllnoqxqxssu`v`xxz{}}llTTȇȉ<<$$  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2UGW2UGW2W2dW2=W2l=W2ǘW2ǘW2W2W2!W2@!W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2klX2 klX2oX2d oX2X2 X2GX2 GX2X2H X2X2hX2Y2DY2a ? IISS]^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI$J$LLN O Q}Q}SSUcVcXXZI[I]]_/`/bebeddfKgKiik1l1nnp\q\ssuBvBxxz({(}}ggډڋMM33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2=W2l=W2ǘW2ǘW2W2W2W2@W2UW2UW2W2W2X2@X2X2 X2X2X2'\X2'\X2bX2 bX2klX2 klX2oX2d oX2X2 X2ݥX2 ݥX2=X2H =X2X2hX2oY2DoY2+$?$IIST^A^Ahhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$h%h'')P*P,,.8/8113 4 6689;|;|==?d@dBBDLELGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpqssuuwqxqz { }}vvZZ̉̋>>""  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X2UX2PUX2sX2,sX2\Y2\Y2 Y2$ Y2Z2Z2[2[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UY2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UY2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UY2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 (H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mD  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)?*?44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33##  ""$q%q'')]*],,.I/I113545668!9!;;= > @@BBDiEiGGIQJQLLN9O9QQS!T!VVX Y [}[}]]_e`ebbdMeMggi5j5llnoqqstvyvyxxza{a}}II11qq b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2CW2TCW2͘W2͘W2W2W2W2(W2IW2IW2W2W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y}6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y}6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y}6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v?Dd b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]6  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2UGW2UGW2W2LW2CW2TCW2͘W2͘W2W2W2'W2('W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2uX20 uX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2)4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 VY2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 VY2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 VY2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 l:Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1rI w] } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO)7*744>?IIST^s^shhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$S%S'');*;,,.#/#113 4 6688:g;g==?M@MBBD1E1GGIJLLNNPkQkSSUOVOXXZ3[3]]_`bbddfmgmiikQlQnnp5q5ssuvxxzz|o}oSSńņ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2CW2TCW2͘W2͘W2W2W2W2(W2OW2OW2W2W2X2(X2X2X2X2X2!\X2!\X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X25X2P5X2X2,X2\Y2\Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y}7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y}7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y}7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 } b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hAs-m } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  LL)2*244>?IISS]i^ihhrKsK} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ooU U""$;%;'')!*!,,./1z1z335`6`88:F;F==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3gg׋׍GG b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2sX20 sX2X2PX2kX2,kX2[Y2[Y2#Y2$#Y2Z2Z25[25[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @[4 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @[4 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @[4  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 *J5U b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ^3Y  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>9?9IIS%T%^^h i s~s~} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMOOQQSmTmVVX[Y[[[]I^I``b5c5eeghjjl m o|o|qqsbtbvvxHyH{{}.~.llތގPP‘“ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2UW2TUW2ߘW2ߘW2W2W2W2(W2CW2CW2'W2'W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2)X2P)X2aX2,aX2w\Y2w\Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 YA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 YA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 YA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`@  0 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2X2 X2X20 X2UX2PUX2Y2,Y2q\Y2q\Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 m b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]2\ . } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbdXeXggi>j>lln"o"qqstvuvuxxzW{W}}99ooQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2UW2TUW2ߘW2ߘW2W2W2W2(W2IW2IW2'W2'W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X21X2P1X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 YB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 YB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 YB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`@  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>#?#IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikhlhnnpLqLssu0v0xxz{}}jj܄܆PPÉË66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2aW2TaW2W2W2!W2!W23W2(3W2W2W23W23W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2CX2PCX2aX2,aX2 \Y2 \Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Iq J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Iq J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Iq J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 8 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1W.  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)C*C44>3?3IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$k%k'')[*[,,.K/K113;4;668+9+;;=>@@B C EEGGIsJsLLN_O_QQSKTKVVX7Y7[[]#^#``bceeggiqjqlln]o]qqsItIvvx5y5{{}!~! oo[[ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2=W2=W2W2W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2_X20 _X2X2PX2aX2,aX2+]Y2+]Y2Y2$Y2Z2Z2/[2/[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y}5 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y}5 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y}5  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 \|$D b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1kVA } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2aW2TaW2W2W2!W2!W23W2(3W2W2W23W23W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2MX20 MX2X2PX2X2,X2\Y2\Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jp I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jp I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jp I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 d  - b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\3 aW } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>'?'IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''uua a""$M%M'')9*9,,.%/%11346688:s;s==?_@_BBDKEKGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//ss[[ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2OW2TOW2٘W2٘W2W2W2W2(W2CW2CW2!W2!W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2kX2 kX2X20 X2 X2P X2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y}6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y}6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p<  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y}6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,mK } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  NN)8*844>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddNN 8 8  ""  kkU U""$?%?''))*),,./11335r6r88:\;\==?F@FBBD0E0GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2eX20 eX2X2PX2SX2,SX2\Y2\Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y""B b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1+ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1bD&0p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln0o0qqstvvxy{y{y}}cc؂؄MM‡‰77!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2OW2TOW2٘W2٘W2W2W2W2(W27W27W2!W2!W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2AX2 AX2X20 X2X2PX2gX2,gX2]Y2]Y2#Y2$#Y2Z2Z25[25[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{a-  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 d b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1vlbP } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJvJvLLN^O^QQSFTFVVX.Y.[[]^``bbdrerggiZjZllnBoBqqs*t*vvxy{{}}nnVVʇʉ>>&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2AX20 AX2X2PX2Y2,Y2a]Y2a]Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Yy b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jL. } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikrlrnnpXqXssu>v>xxz${$}} }}cc։֋II// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2OW2TOW2٘W2٘W2W2W2W2(W27W27W2!W2!W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2Y2,Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1XND:F } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)@*@44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$l%l'')T*T,,.>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2OW2TOW2٘W2٘W2W2W2?W2(?W2W2W2!W2!W2X2(X2X2X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2mX2,mX2s]Y2s]Y2#Y2$#Y2Z2Z25[25[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 l b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mZG  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2OW2TOW2٘W2٘W2W2W2W2(W21W21W2!W2!W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2+X2 +X2X20 X2X2PX2mX2,mX2^Y2^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1pV"  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb@` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>$?$IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?P@PBBD6E6GGIJLLNOQuQuSSU[V[XXZA[A]]_'`'bbd e ggiikflfnnpLqLssu2v2xxz{}}qqWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2Y2,Y2!^Y2!^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14!{ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>!?!IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?Q@QBBD5E5GGIJLLNNPoQoSSUSVSXXZ7[7]]_`bbddfqgqiikUlUnnp9q9ssuvxxz{}s}sWWɄɆ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2OW2TOW2٘W2٘W2W2W2W2(W21W21W2!W2!W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2X20 X2UX2PUX2Y2,Y2S_Y2S_Y2Y2$Y2Z2Z2[2[2O\2O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y~2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y~2 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y~2  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ]}  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  PP)6*644>?IISS]m^mhhrOsO} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++ssY Y""$?%?'')%*%,,. / 1~1~335d6d88:J;J==?.@.BBDEGGIIKcLcNNPEQESSU'V'XXZ [ ]z]z__a\b\ddf>g>iik l nnpqssssuuwUxUzz|7}7kkۋۍKK b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2UW2TUW2ߘW2ߘW2W2W2EW2(EW2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2;X2,;X2'^Y2'^Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Z3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Z3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Z3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  -c2R b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xph } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2UW2TUW2ߘW2ߘW2W2W2W2(W2%W2%W2'W2'W2X2(X2X2X2YX2YX2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2UX2PUX2mX2,mX2#_Y2#_Y2#Y2$#Y2Z2Z25[25[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1eK  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 5U b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  WW)A*A44>,?,IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::** v v""$`%`'')J*J,,.4/411346689;};}==?g@gBBDQEQGGI1?1IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd\e\ggiBjBlln&o&qqstvyvyxxz[{[}}==ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2YW2LYW2UW2TUW2ߘW2ߘW2W2W2W2(W2+W2+W2'W2'W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2qX2 qX2X20 X2X2PX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y3 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y3 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y3  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1;{ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(2<FZ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  SS)=*=44>'?'IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,xxb b""$L%L'')6*6,,. / 113 4 6688:i;i==?S@SBBD=E=GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnTTljNj::  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2gW2$gW2]W2]W2W2W29W2$9W2/X2/X2[X2[X2 `X2$ `X2X2X2X2X23X2 3X2UNY2| UNY2MY2 MY2%Z2 %Z2_[2 _[2P\2 P\2"]2"]215]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  hh)X*X44>F?FIIS2T2^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88((  "#%%'')p*p,,.`/`113P4P668@9@;;=0>0@@B C EEGHJJLLNrOrQQS^T^VVXJYJ[[]6^6``b"c"eeghjjllnpopqqs\t\vvxHyH{{}4~4  nn b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2X2`X2X2X23X23X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y?  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y?  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y?  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>=?=IIS+T+^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIeJeLLNSOSQQSATAVVX/Y/[[]^``b c eeggipjpllnYoYqqs=t=vvx!y!{{}~ww[[͇͉??## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2gW2$gW2WW2WW2W2W29W2$9W2)X2)X2[X2[X2 `X2$ `X2X2X2X2X23X2 3X2UNY2| UNY25Y2 5Y2 Z2  Z2G[2 G[2AP\2 AP\2"]2"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  bb)N*N44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&t t""$`%`'')L*L,,.8/8113$4$6689;;==?r@rBBD^E^GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((ll b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2SW2 SW2W2W2W2W2%W2%W2W2W2YX2pYX2[X2[X2bX2bX2uX2`uX2X2X23X23X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1YA  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1YA  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1YA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  ]])G*G44>1?1IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss]] G G  11zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUrVrXXZ\[\]]_F`Fbbd0e0ggijllnoqsqsssuWvWxxz;{;}}uuYYˎː== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2aW2aW2WW2WW2W2W23W23W2)X2p)X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdkekggiUjUlln?o?qqs)t)vvxy{{}}rr\\чщFF00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2]X2`]X2X2X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bH b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bH b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bH b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  gg)S*S44>=?=IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC77++  "#%%' ( **,,.m/m113Y4Y668E9E;;=1>1@@BCEEG H J}J}LLNeOeQQSMTMVVX5Y5[[]^``bceyeyggiajallnIoIqqs1t1vvxy{{}~uu]]чщEE-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2aW2aW2QW2QW2W2W23W23W2#X2p#X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2)Y2 )Y2Z2p Z2;[2t ;[2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)I*I44>1?1IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI5J5LLNOQQSTVtVtXXZZ[Z]]_@`@bbd&e&ggi j llnnpeqessuKvKxxz1{1}}ppVVɎɐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bJ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bJ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1bJ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)K*K44>1?1IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33##  "#%w%w'')_*_,,.G/G113/4/6689;;==?s@sBBD[E[GGIAJALLN%O%QQS T V{V{XXZ_[_]]_C`Cbbd'e'ggi j l}l}nnpaqassuEvExxz){)}} ccՉՋGG++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2gW2gW2QW2QW2W2W29W29W2#X2p#X2[X2[X2 `X2 `X2X2`X2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)P*P44>8?8IIS T ^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD44   " # %%'')n*n,,.Z/Z113F4F668292;;=>@@BCExExGGI`J`LLNHOHQQS0T0VVXY[[]^`t`tbbd\e\ggiDjDlln,o,qqstvvxxzp{p}}XX̂̄@@%% ~~ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2ݎW2ݎW2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2X2`X2{X2{X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kQ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kQ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kQ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1} } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.N  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  WW)A*A44>'?'IIS T ^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llXX D D  00||f f""$P%P''):*:,,.$/$11346688:m;m==?S@SBBD9E9GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiikilinnpOqOssu5v5xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2gW2gW2KW2KW2W2W29W29W2X2pX2[X2[X2 `X2 `X2]X2`]X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ^^)F*F44>*?*IIST^^hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22 z z""$b%b'')J*J,,.2/211346689;v;v==?Z@ZBBD>E>GGI"J"LLNOQxQxSSU\V\XXZ@[@]]_$`$bbdegzgziik^l^nnpBqBssu&v&xxz { }|}|``҄҆BB$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2ݎW2ݎW2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2[2t [2O\2T O\2W!]2W!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kS b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kS b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kS b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1} } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  WW)=*=44>?IIST^r^rhhrRsR} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22zz` `""$F%F''),*,,,./11335k6k88:Q;Q==?3@3BBDEGGIIKhLhNNPJQJSSU,V,XXZ[]]__aabaddfCgCiik%l%nnpqsxsxuuwZxZzz|<}<nnދލNN b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2[W2[W2EW2EW2W2W2-W2-W2X2pX2[X2[X2_X2_X2QX2`QX2X2X2'X2'X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1f  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  bb)T*T44>D?DIIS0T0^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppbb T T  FF88**  "#%%'(*y*y,,.k/k113]4]668O9O;;=A>A@@B3C3EEG%H%JJLMOOQRTxTxVVXfYf[[]T^T``b@c@eeg*h*jjlmooqqsmtmvvxSyS{{}9~9uuYYˑ˓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tZ& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tZ& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1tZ& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  XX)B*B44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++ w w""$a%a'')K*K,,.5/51134668 9 ;~;~==?h@hBBDRERGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33yycc b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2[W2[W2EW2EW2W2W2-W2-W2X2pX2[X2[X2_X2_X2cX2`cX2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIST^^hispsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdaeaggiGjGlln+o+qqs t v~v~xxz`{`}}BB$$vvXX b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t\( b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t\( b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t\( b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  VV)@*@44>(?(IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //{{e e""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI(J(LLNOQQSSUgVgXXZM[M]]_3`3bbdeggiikmlmnnpQqQssu5v5xxz{}}ooSSƉƋ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2gW2gW2?W2?W2W2W29W29W2X2pX2[X2[X2 `X2 `X2EX2`EX2X2X2!X2!X2INY2L INY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  \\)L*L44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,,   ""$t%t'')d*d,,.T/T113D4D668494;;=$>$@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2gW2gW2?W2?W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  VV)B*B44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..||h h""$T%T'')@*@,,.,/,11346689;z;z==?f@fBBDRERGGI%?%IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLNOQ|Q|SSUfVfXXZP[P]]_:`:bbd$e$ggijllnnpgqgssuKvKxxz/{/}}iiۉۋMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2OW2OW29W29W2W2W2!W2!W2 X2p X2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2X2`X2oX2oX2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations11Q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIqJqLLNYOYQQSATAVVX)Y)[[]^``bbdmemggiUjUlln=o=qqs%t%vvx y {{}}ii݂݄QQŇʼn99!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2OW2OW23W23W2W2W2!W2!W2X2pX2[X2[X2_X2_X2cX2`cX2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)=*=44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n: b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>%?%IIS T ^{^{hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI5J5LLNOQQSSUoVoXXZS[S]]_7`7bbdeggiikqlqnnpUqUssu9v9xxz{}}ssWWɉɋ;; b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W23W23W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2#[2t #[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)D*D44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDlElGGITJTLLN?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?G@GBBD-E-GGIJLLNNPlQlSSURVRXXZ8[8]]_`bbdeggiikglgnnpMqMssu3v3xxz{}}rrVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W2-W2-W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR):*:44>?IIST^x^xhhrZsZ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&n n""$V%V'')>*>,,.&/&11346688:j;j==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2ˎW2ˎW2W2W2CW2CW2W2W2wX2pwX2\X2\X2obX2obX2?X2`?X2X2X2X2X2=NY2L =NY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1wC b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1} } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]n^nhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?'@'BBD E GzGzIIK\L\NNP>Q>SSU V XXZ[]s]s__a]b]ddf?g?iik!l!nnpqststuuwVxVzz|8}8jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2[GW2[GW2W2 W2IW2IW2-W2-W2W2W2W2W2X2pX2[X2[X2_X2_X2٣X2`٣X29X29X2X2X2=NY2T =NY2)Y2 )Y2QZ2x QZ2/[2| /[2P\2\ P\2y ]2y ]2!3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1S  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%}%}'')o*o,,.a/a113S4S668E9E;;=7>7@@B)C)EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m ooqqsotovvxUyU{{};~;!!ww[[͓͑ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2}X2`}X2X2X2X2X2=NY2L =NY2Y2 Y2?Z2p ?Z2[2t [2#P\2T #P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  LL)@*@44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //  k k""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`MbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%%mm b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2UGW2UGW2W2 W2IW2IW2'W2'W2W2W2W2W2X2pX2[X2[X2_X2_X2/X2`/X2X2X2X2X2=NY2T =NY2Y2 Y29Z2x 9Z2[2| [2P\2\ P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1t  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)B*B44>.?.IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln/o/qqstvvxxzd{d}}FF(( zz\\ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2}X2`}X2X2X2X2X2=NY2L =NY2 Y2  Y23Z2p 3Z2[2t [2O\2T O\2]2]2y2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  JJ)<*<44>$?$IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBDY2\>Y2]^Y2]^Y2[iY2[iY2Y28Y2Z2Z2/[2/[2[2[2GP\2lGP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1  bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1   bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[6  : b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d[6lG"}Xp3K& i   D   z U 0  K>A bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  )D*D44>4?4IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  ((==+ +""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjglln\o\qqs@t@vvx$y${{}~zz^^ЇЉBB&&# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2=W2l=W2ǘW2ǘW2W2W2W2@W2[W2[W2W2W2X2@X2X2 X2X2X2-\X2-\X2bX2 bX2klX2 klX2oX2d oX2X2 X2X2 X2eX2H eX2X2hX2[ Y2D[ Y2>Y2>Y2]^Y2]^Y2[iY2 [iY2)Y2)Y2Z2pZ2;[2t;[2[2T[2P\2P\2_"]2$_"]25]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [i6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [i6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [i6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r44T b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[6)i i  D b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d[6lG"}X,3q6{i  D   z U 0 ^  KA bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS)E*E44>5?5IIS!T!^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$/%/'')*,,./113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjsllnkokqqsWtWvvxCyC{{}/~/}}iiߑߓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2 GW2 GW2W2|W2W2W2KW2KW2ՙW2|ՙW2 W2H W2߰W2߰W2}W2P}W2W2XW2W2W2X2PX2X2X2X2X2O[X2$ O[X2`X2, `X2cX2 cX2ymX2$ ymX2pX2 pX2X2| X2MX2XMX2X2X2X2X2 Y2 Y2e>Y2\e>Y2^Y2^Y2iY2iY2 Y28 Y2Z2Z2[2[2[2[2O\2lO\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1  bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1   bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1`@<| U  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d`@ `@ ^<  x \  V 7  4    m p P0,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^    ):*:44>&?&IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^HH 2 2  {{&& ""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnoqtqtssuXvXxxz<{<}} vvZZ̎̐>># b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2=W2l=W2ǘW2ǘW2W2W2ñW2@ñW2aW2aW2W2W2X2@X2X2 X2X2X23\X23\X2bX2 bX2klX2 klX2oX2d oX2X2 X2X2 X2[X2H [X2X2hX2Y2DY2)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$%''))+r,r..01346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2[GW2[GW2W2dW2CW2lCW2͘W2͘W2W2W2'W2@'W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2d oX2X2 X2X2 X2AX2H AX2X2hX2Y2DY2C.?.IISS]^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbVV J J  >>22&&  "#%%'')p*p,,.\/\113H4H668494;;= > @@B C EEGGInJnLLNVOVQQS>T>VVX&Y&[[]^``bbd,e,ggijllnDoDqqs,t,vvxy{{}}ppXẊ̉@@((  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2CW2lCW2͘W2͘W2W2W2W2@W2UW2UW2W2W2X2@X2X2 X2X2X2'\X2'\X2bX2 bX2qlX2 qlX2oX2d oX2X2 X2X2 X2qX2H qX2X2hX2 Y2D Y2>Y2>Y2]^Y2]^Y2[iY2 [iY2Y2Y2Z2pZ2)[2t)[2[2T[2MP\2MP\2"]2$"]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u::Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[x  =   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d[6lG"}Xp3<i T  D   z U 0  K ,A bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>-?-IIST^P^Phisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFllnnpiqissuSvSxxz{{{}}eeڂڄOOćĉ99##  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2UGW2UGW2W2dW2CW2lCW2͘W2͘W2W2W2'W2@'W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2d oX2X2 X2YX2 YX2X2H X2X2hX2! Y2D! Y2>Y2>Y2]^Y2]^Y2[iY2 [iY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2M4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[>~?   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d[6lG"}X3i  D C    z u U N 0  KAr bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  RR):*:44>"?"IISS]^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22"" z z""$f%f'')N*N,,.6/611346689;z;z==?b@bBBDJEJGGI2J2LLNOQQSSUlVlXXZP[P]]_4`4bgbgddfKgKiik/l/nnpZqZssu>v>xxz"{"}}xx\\Ή΋@@$$  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2CW2lCW2͘W2͘W2W2W2W2@W2[W2[W2W2W2X2@X2X2 X2X2X2-\X2-\X2bX2 bX2qlX2 qlX2oX2d oX2X2 X2X2 X2X2H X2aX2haX2 Y2D Y2>Y2>Y2]^Y2]^Y2[iY2 [iY2Y2Y2Z2pZ2[2t[2[2T[2O\2O\2!]2$!]2_4]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[G" I   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d[6lG"}X3i  D   l z 8 U  0  KxA bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844>"?"IIST^@^@hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD?IIST^u^uhhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##ooY Y""$C%C'')-*-,,./11346v6v88:`;`==?H@HBBD.E.GGIJLLNNPmQmSSUSVSXXZ9[9]]_`bbdegxgxiik^l^nnpDqDssu*v*xxz{}}ii܄܆OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2IW2TIW2ӘW2ӘW2 W2 W2W2(W2OW2OW2W2W2X2(X2X2X2X2X2!\X2!\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2=X2P=X2{Y2,{Y27]Y27]Y2Y2$Y2Z2Z2#[2#[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [{: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [{: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [{:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HvM  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDmEmGGIUJULLN=O=QQS%T%VVX Y [[]]_i`ibbdQeQggi9j9lln!o!qqs t v}v}xxze{e}}MM55uu b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2[GW2[GW2W2LW2CW2TCW2͘W2͘W2W2W2-W2(-W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2qlX2 qlX2oX2L oX2X2 X2)X2 )X2X20 X2X2PX2'Y2,'Y2]Y2]Y2Y2$Y2Z2Z2[2[2oO\2oO\2?!]2?!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V[6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V[6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V[6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .NSs b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(  V - I } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  JJ)0*044>?IISS]g^ghhrIsI} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%mmS S""$9%9'')*,,./1x1x335^6^88:D;D==?(@(BBD E G{G{IIK]L]NNP?Q?SSU!V!XXZ[]t]t__aVbVddf8g8iiklnnpprmsmuuwOxOzz|1}1eeՋՍEE b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2IW2TIW2ӘW2ӘW2 W2 W2W2(W2UW2UW2W2W2X2(X2X2X2X2X2'\X2'\X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2AX2PAX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2 [2 [2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [{; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [{; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [{;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$ { R ) I } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2˥X2 ˥X2+X20 +X2mX2PmX2Y2,Y2]Y2]Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @]8 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @]8 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @]8  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~,L  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations15u  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1/Y.  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>$?$IIS T ^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hbbd0e0ggijllnoqtqtssu\v\xxzD{D}},,tt^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2[W2T[W2W2W2W2W2W2(W2IW2IW2-W2-W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2%X2P%X2Y2,Y2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [E bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [E bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [E  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 h b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1H bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 d F x } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  YY)K*K44>=?=IIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggYY K K  ==//!!  "#%~%~'')p*p,,.b/b113T4T668F9F;;=8>8@@B*C*EEGHJJL M OOQQSqTqVVX_Y_[[]M^M``b9c9eeg#h#jjl m ooqqsftfvvxLyL{{}2~2ppTTƑƓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2-Y2,-Y2]Y2]Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 X -  V } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>!?!IIST^x^xhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,bbdeggiikflfnnpJqJssu.v.xxz{}}hhڄچNN44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2[W2T[W2W2W2W2W2W2(W2OW2OW2-W2-W2X2(X2X2X2X2X2!\X2!\X2bX2 bX2lX2 lX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [F bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [F bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [F  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bbb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1~  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ( l N Z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2gW2TgW2W2W2'W2'W29W2(9W2W2W29W29W2X2(X2X2X2 X2 X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X2CX2PCX2Y2,Y27]Y27]Y2Y2$Y2Z2Z2/[2/[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IsN bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IsN bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IsN  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w> b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1.n bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18jC  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)>*>44>.?.IIST^^hisrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$[%['')I*I,,.7/7113%4%6689;;=>@x@xBBDfEfGGIVJVLLNDODQQS2T2VVX Y [[]^``bbdsesggiajallnJoJqqs.t.vvxy{{}}hhڂڄLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2OW2TOW2٘W2٘W2W2W2W2(W2CW2CW2!W2!W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2YX2 YX2X20 X2X2PX2sY2,sY2^Y2^Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [{9 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [{9 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p<  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [{9  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xeR ? >~ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$o%o'')_*_,,.O/O113?4?668/9/;;=>@@BCEEGGIwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2gW2TgW2W2W2'W2'W29W2(9W2W2W29W29W2X2(X2X2X2 X2 X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2YX2 YX2X20 X2X2PX2Y2,Y2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jr M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jr M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jr M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 uu b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 u N ' G } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2OW2TOW2٘W2٘W2W2W2W2(W2CW2CW2!W2!W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2X20 X2CX2PCX2 Y2, Y2w_Y2w_Y2Y2$Y2Z2Z2)[2)[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [{: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [{: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [{:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,   s+ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W2?W2(?W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2qX20 qX2X2PX2Y2,Y2^Y2^Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p **J b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/o  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tX < t } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  ZZ)F*F44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66**  "#%%'')t*t,,.`/`113L4L668898;;=$>$@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln>o>qqs&t&vvxy{{}}jjނބRRƇƉ::"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2UW2TUW2ߘW2ߘW2W2W2W2(W2=W2=W2'W2'W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2Y2,Y2_Y2_Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}e1  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1t bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PH @ 8 D } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  UU)C*C44>1?1IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIYJYLLNCOCQQS-T-VVXY[[]^`v`vbbd`e`ggiJjJlln4o4qqstvvxy{}{}}}gg܂܄QQƇƉ;;%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W29W2(9W2W2W2'W2'W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X27X2P7X29Y2,9Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 j b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  h L X } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2UW2TUW2ߘW2ߘW2W2W2W2(W2=W2=W2'W2'W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2Y2,Y2I`Y2I`Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 m b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1,l  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 0 (  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2UW2TUW2ߘW2ߘW2W2W2EW2(EW2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2Y2,Y2^Y2^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hWF 5 :z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844> ? IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbd e g|g|iikblbnnpHqHssu.v.xxz{}}mmSSʼnŋ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2UW2TUW2ߘW2ߘW2W2W2W2(W21W21W2'W2'W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2QX2PQX2Y2,Y2[`Y2[`Y2#Y2$#Y2Z2Z25[25[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [}6 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [}6 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1rZ&  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [}6  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Gg b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 Y  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(+ . 1   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)G*G44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;++  "#%y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdUeUggi=j=lln%o%qqs t vvxxzi{i}}QQłń99 yy b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2UW2TUW2ߘW2ߘW2W2W2?W2(?W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2/X2 /X2X20 X2X2PX2KY2,KY2}_Y2}_Y2Y2$Y2Z2Z2[2[2O\2O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $D_Ii b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$   q) } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))q q""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbdegsgsiikWlWnnp;q;ssuvxxz{}u}uYY˄ˆ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2wX20 wX2X2PX2Y2,Y2_Y2_Y2#Y2$#Y2Z2Z25[25[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ,b1Q b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations19y  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@: 4 . @ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  SS)=*=44>(?(IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&&r r""$\%\'')F*F,,.0/011346689;y;y==?c@cBBDMEMGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00xxbb b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2[W2T[W2W2W2W2W2W2(W2+W2+W2-W2-W2X2(X2X2X2_X2_X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2Y2,Y2'aY2'aY2;Y2$;Y2Z2Z2M[2M[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1gO  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Cc b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1   *  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  ]])O*O44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk]] O O  AA33%%  " # %%'')t*t,,.f/f113X4X668J9J;;=<><@@B.C.EEG H JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2;X2 ;X2X20 X2X2PX2cY2,cY2O`Y2O`Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Tt b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 3  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ( "  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2[W2T[W2W2W2W2W2W2(W21W21W2-W2-W2X2(X2X2X2eX2eX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X27X2P7X2Y2,Y2aY2aY2#Y2$#Y2Z2Z25[25[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [7 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [7 bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [7  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 'G b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1n  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 8 FTb^ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>3?3IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!uuWW b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2/W2gW2$gW2cW2cW2W2W29W2$9W25X25X2[X2[X2 `X2$ `X2X2X2uX2uX23X2 3X2[NY2| [NY2AY2 AY2Z2 Z2S[2 S[2kP\2 kP\2;"]2;"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1"B  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>A?AIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIiJiLLNWOWQQSETEVVX3Y3[[]!^!``bceeggitjtlln]o]qqsAtAvvx%y%{{} ~ {{__чщCC'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2iX2`iX2X2X23X23X2[NY2L [NY2MY2 MY2%Z2p %Z2_[2t _[2P\2T P\2"]2"]2+5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[C b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[C b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[C b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  dd)T*T44>B?BIIS.T.^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttdd T T  DD44$$  "#%|%|'')l*l,,.\/\113L4L668<9<;;=,>,@@BCEEG H JJLLNnOnQQSZTZVVXFYF[[]2^2``bceeg h jjllnlolqqsXtXvvxDyD{{}0~0~~jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2)W2gW2$gW2]W2]W2W2W29W2$9W2/X2/X2[X2[X2 `X2$ `X2X2X2X2X23X2 3X2[NY2| [NY2/Y2 /Y2Z2 Z2A[2 A[2P\2 P\2!]2!]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  __)I*I44>3?3IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uu__ I I  33||f f""$P%P''):*:,,.$/$11346688:m;m==?W@WBBDAEAGGI+J+LLNOQQSSUtVtXXZ^[^]]_H`Hbbd2e2ggijllnoququssuYvYxxz={=}}!!ww[[͎͐?? b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2YW2 YW2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2oX2`oX2X2X23X23X2[NY2L [NY25Y2 5Y2 Z2p  Z2G[2t G[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[E b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[E b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1[E b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1|  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)L*L44>6?6IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$^%^'')J*J,,.6/6113"4"6689;;==?p@pBBD\E\GGIFJFLLN.O.QQSTVVXXZr[r]]_Z`ZbbdBeBggi*j*llnoqqssunvnxxzV{V}}>>&&jj b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2aW2aW2]W2]W2W2W23W23W2/X2p/X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  kk)W*W44>A?AIIS)T)^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwkk _ _  SSGG;;//  "###%%'(**,,.q/q113]4]668I9I;;=5>5@@B!C!EEG H JJLLNiOiQQSQTQVVX9Y9[[]!^!``b c e}e}ggiejellnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2W2W2W2W2+W2+W2W2W2_X2p_X2[X2[X2bX2bX2]X2`]X2X2X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dL b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dL b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dL b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1l  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ^^)L*L44>8?8IIS"T"^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp^^ L L  ::(( { {""$i%i'')W*W,,.E/E113343668!9!;;=>@@BBDtEtGGI`J`LLNJOJQQS4T4VVXY[[]^`}`}bbdgegggiQjQlln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2aW2aW2WW2WW2W2W23W23W2)X2p)X2[X2[X2`X2`X2X2`X2cX2cX2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)M*M44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%y%y'')a*a,,.I/I1131416689;;=>@u@uBBD]E]GGICJCLLN'O'QQS T V}V}XXZa[a]]_E`Ebbd)e)ggi j llnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2)Y2 )Y2Z2p Z2;[2t ;[2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dN b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dN b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1dN b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ]])G*G44>/?/IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  o o""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd$e$ggi j l}l}nnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2gW2gW2QW2QW2W2W29W29W2#X2p#X2[X2[X2 `X2 `X2X2`X2X2X2-X2-X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  [[)E*E44>+?+IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    j j""$T%T'')>*>,,.(/(11346688:q;q==?W@WBBD=E=GGI#J#LLN O Q|Q|SSUbVbXXZH[H]]_.`.bbdeggiikmlmnnpSqSssu9v9xxz{}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2׎W2׎W2W2W21W21W2W2W2eX2peX2\X2\X2{bX2{bX2X2`X2X2X2-X2-X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2GP\2T GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mU! b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mU! b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mU! b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  ``)L*L44>4?4IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ppdd X X  LL@@00  "#%~%~'')j*j,,.V/V113B4B668.9.;;=>@@BCEtEtGGI\J\LLNDODQQS,T,VVXY[[]]_p`pbbdXeXggi@j@lln(o(qqstvvxxzl{l}}TTȂȄ<<!!zz b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2gW2gW2QW2QW2W2W29W29W2#X2p#X2[X2[X2 `X2 `X2_X2`_X2X2X2-X2-X2ONY2L ONY2 Y2  Y2Z2p Z2[2t [2O\2T O\2]!]2]!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  YY)?*?44>!?!IIST^t^thhrTsT} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    ||b b""$H%H'').*.,,./11335m6m88:S;S==?5@5BBDEGGIIKjLjNNPLQLSSU.V.XXZ[]]__acbcddfEgEiik'l'nnp q szszuuw\x\zz|>}> ppPP b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2׎W2׎W2W2W27W27W2W2W2kX2pkX2 \X2 \X2{bX2{bX2X2`X2X2X2-X2-X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mW# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mW# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1mW# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)D*D44>(?(IIS T ^~^~hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp`` P P  @@00 x x""$`%`'')H*H,,.0/011346689;t;t==?X@XBBD[>]]_"`"bbdegxgxiik\l\nnp@q@ssu$v$xxz{}z}z^^ЄІ@@"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2[W2[W2KW2KW2W2W2-W2-W2X2pX2[X2[X2_X2_X2SX2`SX2X2X2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  \\)F*F44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// { {""$e%e'')O*O,,.9/9113#4#668 9 ;;==?l@lBBDVEVGGI?J?LLN'O'QQSTVVXXZk[k]]_S`Sbbd;e;ggi#j#lln o qqssugvgxxzO{O}}77}}gg b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v^* b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ^^)P*P44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll^^ P P  BB44&&  " # %%'')u*u,,.g/g113Y4Y668K9K;;==>=@@B/C/EEG!H!JJLMOOQQStTtVVXbYb[[]P^P``b*?*IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUȉȋ;;!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v`, b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v`, b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v`, b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ZZ)H*H44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1llZZ H H  66$$ w w""$e%e'')S*S,,.A/A113/4/6689;;= > @@BBDpEpGGI\J\LLNFOFQQS0T0VVXY[[]^`y`ybbd_e_ggiEjElln)o)qqs t v|v|xxz^{^}}@@""ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2gW2gW2EW2EW2W2W29W29W2X2pX2[X2[X2 `X2 `X2KX2`KX2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1h  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2юW2юW2W2W2=W2=W2W2W2qX2pqX2\X2\X2ubX2ubX2X2`X2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1|} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1|} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g3 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1| } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  XX)H*H44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((  ""$p%p'')`*`,,.P/P113@4@668090;;= > @@BCEEGHJvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2gW2gW2?W2?W2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  SS)=*=44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiSS = =  ''ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?K@KBBD5E5GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2UW2UW2?W2?W2W2W2'W2'W2X2pX2[X2[X2_X2_X25X2`5X2X2X2!X2!X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2YP\2T YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Tt  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  __)K*K44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk__ S S  GG;;//##  "#%%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2X2`X2 X2 X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p< b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>,?,IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln/o/qqstvvxy{x{x}}bbׂׄLL66  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2UW2UW29W29W2W2W2'W2'W2 X2p X2[X2[X2_X2_X2iX2`iX2X2X2!X2!X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)A*A44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2X2`X2 X2 X2!X2!X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r> b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpWqWssu=v=xxz#{#}} ||bbՉՋHH.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2[W2[W29W29W2W2W2-W2-W2 X2p X2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>?IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2ˎW2ˎW2W2W2IW2IW2W2W2}X2p}X2\X2\X2obX2obX2-X2`-X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yE b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yE b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1yE b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>(?(IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;= > @@BBDhEhGGIPJPLLN8O8QQS T VVXY[|[|]]_d`dbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2aGW2aGW2W2 W2[W2[W23W23W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  MM)3*344>?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((ppV V""$<%<'')"*",,./1{1{335a6a88:G;G==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2ˎW2ˎW2W2W2IW2IW2W2W2}X2p}X2\X2\X2obX2obX2EX2`EX2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1{G b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1j  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?L@LBBD0E0GGIJLLNNPjQjSSUNVNXXZ2[2]]_`bbdegtgtiikXlXnnp%?%IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2}W2 }W2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2/Y2 /Y2WZ2p WZ25[2t 5[2P\2T P\2 ]2 ]2'3]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1N b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  RR)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``RR D D  66((   ""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2[GW2[GW2W2 W2IW2IW2-W2-W2W2W2W2W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2T =NY2 Y2  Y23Z2x 3Z2[2| [2O\2\ O\2]2]2y2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1s^  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  LL):*:44>"?"IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qq[ [""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2}W2 }W2W2W2W2W2OW2OW2W2W2X2pX2!\X2!\X2bX2bX2X2`X2!X2!X2X2X2=NY2L =NY2Y2 Y2?Z2p ?Z2[2t [2P\2T P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1   b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)D*D44>0?0IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2W2W2KW2KW2ՙW2|ՙW2 W2H W2W2W2W2PW2W2XW2W2W2X2PX2X2X2X2X2U[X2$ U[X2`X2, `X2cX2 cX2ymX2$ ymX2pX2 pX2X2| X2GX2XGX2X2X2X2X2uY2uY21@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``RR D D  66((SSE E""$%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b3?3IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55))  "#%I%I'')5*5,,.!/!113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWllnKoKqqs3t3vvxy{{}~ww__ӇӉGG//  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2W2W2KW2KW2ՙW2|ՙW2 W2H W2߰W2߰W2}W2P}W2W2XW2W2W2X2PX2X2X2X2X2O[X2$ O[X2`X2, `X2cX2 cX2ymX2$ ymX2pX2 pX2X2| X2eX2XeX2X2X2X2X2Y2Y2[2?2IIST^^hiszsz} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **<<* *""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln3o3qqstvvxxzh{h}}JJ,,bb# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2CW2lCW2͘W2͘W2W2W2ɱW2@ɱW2gW2gW2W2W2X2@X2X2 X2X2X29\X29\X2bX2 bX2qlX2 qlX2oX2d oX2X2 X2X2 X2[X2H [X2X2hX2Y2DY2'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$*%*'')*,,..013#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II--  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2aGW2aGW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2-W2@-W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2MX2 MX2X2H X2X2hX2{Y2D{Y270?0IISS]^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``PP @ @  00   x x""$h%h'')X*X,,.H/H113848668(9(;;=>@@BCEEGGIpJpLLN\O\QQSHTHVVX4Y4[[] ^ ``bbdBeBggi.j.llndodqqsPtPvvx?IISS]z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?K@KBBD1E1GGIJLLNNPpQpSSUVVVXXZ<[<]]_"`"bWbWddf=g=iik#l#nnpIqIssu/v/xxz{}}nnTTƉƋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2[GW2[GW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2-W2@-W2W2W2W2W2X2@X2X2 X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2qX2 qX2X2H X2X2hX2a Y2Da Y2>Y2>Y2^Y2^Y2iY2 iY2 Y2 Y2Z2pZ2[2t[2[2T[2P\2P\2!]2$!]24]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1] b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1] g  D b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d]:t)Qr.; hE"  \  9   s P  ^ -m gD bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  LL)8*844>$?$IISS]^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$rr^ ^""$J%J'')6*6,,."/"11346688:p;p==?\@\BBDHEHGGI4J4LLNOQQSTVxVxXXZ`[`]]_H`Hb{b{ddfcgciikKlKnoq|q|ssudvdxxzL{L}}44xx``  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2ñW2@ñW2aW2aW2W2W2X2@X2X2 X2X2X23\X23\X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2X2 X2X2H X2/X2h/X2Y2DY2O?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__abddffhmimkkmnpqsrsruuwTxTzz|6}6jjڋڍJJ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2aX2,aX2ZY2ZY2Y2$Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U]: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U]: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U]:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1IWr  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  NN)<*<44>*?*IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``NN < <  **}}k k""$Y%Y'')G*G,,.5/5113#4#6689;;==?v@vBBDdEdGGIRJRLLN bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]y> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]y>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 3S b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Y2   } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>'?'IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI7J7LLNOQQSTV{V{XXZc[c]]_K`Kbbd3e3ggijllnoqwqwssu_v_xxzG{G}}//wwaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2aGW2aGW2W2LW2IW2TIW2ӘW2ӘW2 W2 W23W2(3W2W2W2W2W2X2(X2X2X2X2X2[X2[X2bX2 bX2wlX2 wlX2oX2L oX2X2 X2wX2 wX2X20 X2X2PX2X2,X2{[Y2{[Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V]: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V]: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V]:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1e,s 9  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)7*744>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz``ӉӋFF,, b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2OW2TOW2٘W2٘W2W2W2W2(W2[W2[W2!W2!W2X2(X2X2X2X2X2-\X2-\X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]y? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]y? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]y?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 kPpu b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d=oY } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2X20 X2%X2P%X2qX2,qX2[Y2[Y2Y2$Y2Z2Z2#[2#[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @_< bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @_< bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @_<  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P p P  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)@*@44>*?*IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddXX L L  @@44$$  ""$r%r'')^*^,,.J/J113646668"9";;=>@@BBDjEjGGIRJRLLN:O:QQS"T"VVX Y [~[~]]_f`fbbdNeNggi6j6llnoqqstvzvzxxzb{b}}JJ22rr b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2[W2T[W2W2W2W2W2W2(W2OW2OW2-W2-W2X2(X2X2X2X2X2!\X2!\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2CX20 CX2X2PX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2/[2/[2GP\2GP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ] I bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ] I bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ] I  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ' b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1dH,0p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIYJYLLNGOGQQS5T5VVX#Y#[[]^``bbdvevggidjdllnMoMqqs1t1vvxy{{}}kk݂݄OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2Y2,Y2)\Y2)\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2/4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 wJjo b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@`0  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IIST^t^thhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bbddfngniikRlRnnp6q6ssuvxxzz|p}pTTƄƆ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2W2(W2OW2OW23W23W2X2(X2X2X2X2X2!\X2!\X2cX2 cX2lX2 lX2oX2L oX2X2 X2{X2 {X2X20 X2X2PX2Y2,Y2[]Y2[]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ] J bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ] J bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ] J  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tX< } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  MM)7*744>#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2mW2TmW2W2W2-W2-W2?W2(?W2W2W2?W2?W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2eX2 eX2X20 X2X2PX2gX2,gX2M\Y2M\Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IuR bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IuR bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IuR  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Pp b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1[6  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  VV)H*H44>:?:IIS&T&^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddVV H H  ::,,  "#%{%{'')m*m,,._/_113Q4Q668C9C;;=5>5@@B'C'EEGHJJL M OOQQSnTnVVX\Y\[[]J^J``b6c6eeg h jjl m o}o}qqsctcvvxIyI{{}/~/mmߌߎQQÑÓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2UW2TUW2ߘW2ߘW2W2W2W2(W2IW2IW2'W2'W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2_X2 _X2X20 X2X2PX2gX2,gX2m]Y2m]Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]y= bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]y= bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]y=  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1o^M  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  ]])I*I44>5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ii]] Q Q  EE99--!!  "#%%'(*w*w,,.c/c113O4O668;9;;;='>'@@BCEEGGIuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnAoAqqs)t)vvxy{{}}mmUUɇɉ==%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2gW2TgW2W2W2'W2'W29W2(9W2W2W29W29W2X2(X2X2X2 X2 X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2 X2  X2kX20 kX2X2PX2Y2,Y2\Y2\Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JtQ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1` b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JtQ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JtQ  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p&&F b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lG"}[ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2UW2TUW2ߘW2ߘW2W2W2W2(W2IW2IW2'W2'W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2UX2PUX2Y2,Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]y> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]y> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]y>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<+  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)?*?44>'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2UW2TUW2ߘW2ߘW2W2W2EW2(EW2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2GX20 GX2X2PX2wX2,wX2]Y2]Y2/Y2$/Y2Z2Z2A[2A[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ii" b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fL2p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  TT)D*D44>4?4IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$ | |""$l%l'')\*\,,.L/L113<4<668,9,;;=>@@B C EEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln^o^qqsJtJvvx6y6{{}"~"pp\\ґғ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2UW2TUW2ߘW2ߘW2W2W2W2(W2=W2=W2'W2'W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2 X20  X2KX2PKX2gX2,gX2!^Y2!^Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i5  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ];  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 (H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ztn } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>#?#IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?O@OBBD5E5GGIJLLNOQtQtSSUZVZXXZ@[@]]_&`&bbd e ggiikelennpKqKssu1v1xxz{}}ppVVȉȋ:: b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2UW2TUW2ߘW2ߘW2W2W2?W2(?W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2Y2,Y2]Y2]Y2Y2$Y2Z2Z2)[2)[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~dJV } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2UW2TUW2ߘW2ߘW2W2W2W2(W2CW2CW2'W2'W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2%Y2,%Y2^Y2^Y2Y2$Y2Z2Z2 [2 [2O\2O\2Q!]2Q!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ];  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 DdIi b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hb\VJ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  OO)5*544>?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2_X2P_X2mX2,mX2]Y2]Y2)Y2$)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1qbS ` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln1o1qqstvvxy{z{z}}ddقلNNÇÉ88"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2[W2T[W2W2W2W2W2W2(W27W27W2-W2-W2X2(X2X2X2kX2kX2 \X2 \X2bX2 bX2lX2 lX2oX2L oX2X2 X2[X2 [X2X20 X2X2PX2mX2,mX2^Y2^Y2#Y2$#Y2Z2Z25[25[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t^* b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]{: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t^* b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]{: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t^*  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]{:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 3SSsx b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 @ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  VV)@*@44>+?+IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99)) u u""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`Obbd7e7ggijllnoq{q{ssucvcxxzK{K}}33{{ee b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2Y2,Y2c^Y2c^Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D5& } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2[W2T[W2W2W2W2W2W2(W2=W2=W2-W2-W2X2(X2X2X2qX2qX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2YX20 YX2X2PX2Y2,Y2_Y2_Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]z: bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]z: bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]z:  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 7W. b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1T bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1#E } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  RR)<*<44>&?&IIS T ^}^}hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++wwa a""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88((  "#%v%v'')b*b,,.N/N113:4:668&9&;;=>@@BBDnEnGGIVJVLLN>O>QQS&T&VVXY[[]]_j`jbbdReRggi:j:lln"o"qqs t v~v~xxzf{f}}NN‚„66vv b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2_W2L_W2aW2TaW2W2W2!W2!W2W2(W21W21W23W23W2X2(X2X2X2eX2eX2\X2\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2?X2P?X2sX2,sX2_Y2_Y2/Y2$/Y2Z2Z2A[2A[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1iS  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ];  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 5 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 00 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  WW)E*E44>5?5IIS#T#^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiWW E E  33!!t t""$b%b'')P*P,,.>/>113,4,6689;;=>@@BBDmEmGGI]J]LLNKOKQQS9T9VVX'Y'[[]^``bcezezggihjhllnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2aW2TaW2W2W2!W2!W2KW2(KW2W2W23W23W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2}X20 }X2X2PX2)Y2,)Y2#_Y2#_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2G4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^;  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .[{3S b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1plhd } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2aW2TaW2W2W2!W2!W2W2(W27W27W23W23W2X2(X2X2X2kX2kX2 \X2 \X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2)X2P)X2CY2,CY2U`Y2U`Y2Y2$Y2Z2Z2/[2/[2O\2O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]; bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]; bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ];  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 *J{ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@P`p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  QQ);*;44>'?'IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggQQ ; ;  %%nnX X""$B%B''),*,,,./11346u6u88:_;_==?I@IBBD3E3GGIJLLN O Q~Q~SSUhVhXXZR[R]]_<`M?MIIS9T9^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1yykk ] ]  OOAA33%%  "#%%' ( **,,.t/t113f4f668X9X;;=J>J@@B@?@IIS(T(^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvjj ^ ^  RRFF::..  ""#"%%'(**,,.p/p113\4\668H9H;;=4>4@@B C EEG H JJLLNhOhQQSPTPVVX8Y8[[] ^ ``bce|e|ggidjdllnLoLqqs4t4vvxy{{}~xx``ԇԉHH00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2/W2gW2$gW2cW2cW2W2W29W2$9W25X25X2[X2[X2 `X2$ `X2ŢX2ŢX2%X2%X23X2 3X2[NY2| [NY25Y2 5Y2 Z2  Z2G[2 G[25P\2 5P\2"]2"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14s b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww  ee)S*S44>???IIS'T'^^h i sysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1wwee S S  AA//   ""$p%p'')^*^,,.L/L113:4:668(9(;;=>@@BCE{E{GGIgJgLLNQOQQQS;T;VVX%Y%[[]^``bbdjejggiPjPlln4o4qqstvvxxzi{i}}KK--aa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2_W2 _W2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2AX2`AX2X2X23X23X2[NY2L [NY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]I b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]I b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1]I b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ll  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  ff)N*N44>4?4IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvff V V  FF66&&  "#%z%z'')b*b,,.J/J1132426689;;=>@v@vBBD^E^GGIDJDLLN(O(QQS T V~V~XXZb[b]]_F`Fbbd*e*ggijllnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2aW2aW2]W2]W2W2W23W23W2/X2p/X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2MY2 MY2%Z2p %Z2_[2t _[2P\2T P\2"]2"]2+5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu  ee)U*U44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1uuee U U  EE55%%  "#%}%}'')m*m,,.]/]113M4M668=9=;;=->-@@BCEEG H JJLLNoOoQQS[T[VVXGYG[[]3^3``bceeg h jjllnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fP b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fP b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fP b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ZZ)D*D44>*?*IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo[[ G G  33  i i""$S%S'')=*=,,.'/'11346688:p;p==?V@VBBD5?5IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ss__ K K  77##q q""$]%]'')I*I,,.5/5113!4!668 9 ;;==?o@oBBD[E[GGIEJELLN-O-QQSTVVXXZq[q]]_Y`YbbdAeAggi)j)llnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2X2`X2X2X2-X2-X2UNY2L UNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2c!]2c!]2 4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fR b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fR b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1fR b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ZZ)@*@44>"?"IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  }}c c""$I%I'')/*/,,./11335n6n88:T;T==?6@6BBDEGGIIKkLkNNPMQMSSU/V/XXZ[]]__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2gW2gW2WW2WW2W2W29W29W2)X2p)X2[X2[X2 `X2 `X2X2`X2X2X2-X2-X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  __)M*M44>9?9IIS#T#^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq__ M M  ;;)) | |""$j%j'')X*X,,.F/F113444668"9";;=>@@BBDuEuGGIaJaLLNKOKQQS5T5VVXY[[] ^ `~`~bbdhehggiRjRlln  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  [[)E*E44>.?.IIST^^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn^^ N N  >>.. z z""$d%d'')N*N,,.8/8113"4"668 9 ;;==?k@kBBDUEUGGI>J>LLN&O&QQSTVVXXZj[j]]_R`Rbbd:e:ggi"j"lln o q~q~ssufvfxxzN{N}}66||ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2mW2mW2WW2WW2W2W2?W2?W2)X2p)X2[X2[X2`X2`X2X2`X2X2X2-X2-X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)F*F44>.?.IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI2J2LLNOQQSSUqVqXXZW[W]]_=`=bbd#e#ggi j l|l|nnpbqbssuHvHxxz.{.}}mmSSƎƐ99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2׎W2׎W2W2W27W27W2W2W2kX2pkX2 \X2 \X2{bX2{bX2X2`X2X2X2-X2-X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o[' b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  YY)C*C44>+?+IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnZZ F F  22  ~~h h""$R%R'')<*<,,.&/&11346688:o;o==?Y@YBBDCECGGI+J+LLNOQQSSUjVjXXZP[P]]_6`6bbdeggiikplpnnpTqTssu8v8xxz{}}rrVVɉɋ<<"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2[W2[W2QW2QW2W2W2-W2-W2#X2p#X2[X2[X2_X2_X2YX2`YX2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  aa)M*M44>5?5IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqee Y Y  MMAA11  " # %%'')k*k,,.W/W113C4C668/9/;;=>@@BCEuEuGGI]J]LLNEOEQQS-T-VVXY[[]]_q`qbbdYeYggiAjAlln)o)qqstvvxxzm{m}}UUɂɄ==""{{ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2)Y2 )Y2Z2p Z2;[2t ;[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xb. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xb. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xb. b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  XX)F*F44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjXX F F  44""u u""$c%c'')Q*Q,,.?/?113-4-6689;;= > @@BBDnEnGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2[W2[W2KW2KW2W2W2-W2-W2X2pX2[X2[X2_X2_X2]X2`]X2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1s  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  [[)C*C44>'?'IIS T ^}^}hhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo__ O O  ??// w w""$_%_'')G*G,,.///11346688:s;s==?W@WBBD;E;GGIJLLNOQuQuSSUYVYXXZ=[=]]_!`!bbdegwgwiik[l[nnp?q?ssu#v#xxz{}y}y]]τφ??!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2X2`X2EX2EX2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xd0 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xd0 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xd0 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  TT)>*>44>(?(IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjTT > >  ((qq[ [""$E%E'')/*/,,./11346x6x88:b;b==?L@LBBD6E6GGI J LLN O QQSSUiViXXZS[S]]_=`=bbd'e'ggijllnnpjqjssuNvNxxz2{2}}llމދPPސ44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2gW2gW2KW2KW2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  __)Q*Q44>A?AIIS-T-^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm__ Q Q  CC55''  " # %%'')v*v,,.h/h113Z4Z668L9L;;=>>>@@B0C0EEG"H"JJLMOOQQSuTuVVXcYc[[]Q^Q``b=c=eeg'h'jjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2юW2юW2W2W2CW2CW2W2W2wX2pwX2\X2\X2ubX2ubX2X2`X2uX2uX2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1z} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1z} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k7 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1z } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations10P  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ^^)J*J44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::..""  "#%%'(*x*x,,.d/d113P4P668<9<;;=(>(@@BCEEGHJtJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2gW2gW2EW2EW2W2W29W29W2X2pX2[X2[X2 `X2 `X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  YY)G*G44>3?3IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkYY G G  55##v v""$d%d'')R*R,,.@/@113.4.6689;;= > @@BBDoEoGGI[J[LLNEOEQQS/T/VVXY[[]^`x`xbbd^e^ggiDjDlln(o(qqs t v{v{xxz]{]}}??!!ssUU b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2юW2юW2W2W2CW2CW2W2W2wX2pwX2\X2\X2ubX2ubX2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1z} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1z} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1z } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>(?(IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI8J8LLNOQQSTVrVrXXZV[V]]_:`:bbdeggijltltnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2UW2UW2EW2EW2W2W2'W2'W2X2pX2[X2[X2_X2_X2KX2`KX2X2X2!X2!X2CNY2L CNY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1h  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii  YY)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1iiYY I I  99))   ""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnaoaqqsMtMvvx9y9{{}%~%ss__ՑՓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2}X2`}X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t@ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>?IIST^w^whhr[s[} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdegzgziik`l`nnpFqFssu,v,xxz{}}kkބކOO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2UW2UW2?W2?W2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS)?*?44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggSS ? ?  ++yye e""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2KX2`KX2X2X2!X2!X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2uO\2T uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vB b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1h  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]i^ihhrIsI} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__aXbXddf:g:iiklnnpprosouuwQxQzz|3}3eeՋՍEE b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>-?-IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIUJULLN?O?QQS)T)VVXY[[]]_r`rbbd\e\ggiFjFlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2}W2 }W2ˎW2ˎW2W2W2IW2IW2W2W2}X2p}X2\X2\X2obX2obX23X2`3X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2#P\2T #P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}I b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}I b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1}I b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  OO)9*944>"?"IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbRR B B  22""n n""$X%X'')B*B,,.,/,11346689;u;u==?_@_BBDIEIGGI2J2LLNOQQSTVvVvXXZ^[^]]_F`Fbbd8e8ggi j llnoq|q|ssudvdxxzL{L}}44zzdd b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2[W2[W29W29W2W2W2-W2-W2 X2p X2[X2[X2_X2_X2oX2`oX2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP):*:44>"?"IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI&J&LLN O QQSSUeVeXXZK[K]]_1`1bbdeggiikplpnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2}W2 }W2ˎW2ˎW2W2W2OW2OW2W2W2X2pX2!\X2!\X2obX2obX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>?IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGIJLLNOQxQxSSU^V^XXZD[D]]_*`*bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2gGW2gGW2W2 W2IW2IW23W23W2W2W2W2W2X2pX2[X2[X2_X2_X2;X2`;X2X2X2X2X2=NY2T =NY2Y2 Y2EZ2x EZ2#[2| #[25P\2\ 5P\2+ ]2+ ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1O  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)C*C44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2OW2OW2W2W2X2pX2!\X2!\X2bX2bX2 X2` X2iX2iX2X2X2=NY2L =NY2#Y2 #Y2KZ2p KZ2)[2t )[2MP\2T MP\2C ]2C ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1R b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1SO  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  LL)D*D44>2?2IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^LL : :  (({{i i""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2aGW2aGW2W2 W2IW2IW23W23W2W2W2W2W2X2pX2[X2[X2_X2_X25X2`5X2X2X2X2X2=NY2T =NY2Y2 Y2-Z2x -Z2 [2|  [2O\2\ O\2]2]2[2]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1#|  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  OO);*;44>?IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##k k""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2UW2UW2W2W2X2pX2'\X2'\X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y29Z2p 9Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1T b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^  HH):*:44>$?$IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1^^HH 2 2  {{eeO O""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2W2W2QW2QW2ۙW2|ۙW2W2HW2W2W2W2PW2W2XW2#W2#W2X2PX2X2X2X2X2U[X2$ U[X2`X2, `X2cX2 cX2mX2$ mX2pX2 pX2X2| X2SX2XSX2X2X2X2X2Y2Y2=Y2d=Y2^Y2^Y2iY2iY2Y2@Y2Z2Z2)[2)[2[2[2#P\2t#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1P  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_>( h h  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:x\W'6r =  t  S  2   p O.,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__    )@*@44>+?+IIST^^his}s}} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //??) )""$_%_'')I*I,,.3/311346689;|;|==?f@fBBDPEPGGI;J;LLN#O#QQS T VVXXZg[g]]_O`ObbdCeCggi+j+llnoqqssuovoxxzW{W}}??''qq# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2ɱW2@ɱW2gW2gW2W2W2X2@X2X2 X2X2X29\X29\X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2X2 X2YX2H YX2X2hX2{Y2D{Y27 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _a> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _a>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |((H b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_>" " b  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:x9W~C6   t  S  2   r s Q0.n bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)@*@44>.?.IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$%'') * ,,..013'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbd]e]ggiGjGlln=o=qqs't'vvxy{{}}ppZZχωDD..  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2W2W2QW2QW2ۙW2|ۙW2W2HW2W2W2W2PW2W2XW2#W2#W2X2PX2X2X2X2X2U[X2$ U[X2`X2, `X2cX2 cX2mX2$ mX2pX2 pX2X2| X2X2XX2MX2MX2X2X2 Y2 Y2=Y2d=Y2g`Y2g`Y2ekY2ekY2 Y2@ Y2Z2Z2[2[2[2[2O\2tO\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1? b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1v,l   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1dvlbXND:0$   S    t i  h  \UP|0p$ bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  )<*<44>&?&IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##,, ""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _b> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _b>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k&&F b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_>k  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:xW6 O  t   S Y  2    p O 8.,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM);*;44>%?%IIS T ^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$ % ''))+j,j..013 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpeqessuKvKxxz1{1}}ppVVɎɐ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2aGW2aGW2W2dW2OW2lOW2٘W2٘W2W2W23W2@3W2W2W2!W2!W2X2@X2X2 X2X2X2[X2[X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2YX2 YX2X2P X2X2pX2Y2LY2=" " b  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:xKW6]&   t  S  2   p O.,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  MM);*;44>+?+IISS]^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__MM ; ;  ))||j j""$X%X'')F*F,,.4/4113"4"6689;;==?u@uBBDcEcGGISJSLLNAOAQQS/T/VVXY[[] ^ ``bbd.e.ggijllnQoQqqs5t5vvxy{{}}ooSSŇʼn77  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2ñW2@ñW2aW2aW2W2W2X2@X2X2 X2X2X23\X23\X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2wX2 wX2X2H X2X2hX2Y2DY2[ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1iRW B  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1diR;$ v_H3`0  t _ J 5  4t I]H bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)B*B44>,?,IISS]^hhrrsr} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffZZ N N  BB66&&  ""$t%t'')`*`,,.L/L113848668$9$;;=>@@BBDlElGGITJTLLN<|  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:xW6 o  t 8  S   2 J   p O.n.,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  II)3*344>?IISS]^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__II 3 3  ||ffP P""$:%:'')$*$,,./11335m6m88:W;W==?A@ABBD+E+GGIJLLNOQvQvSSU`V`XXZJ[J]]_4`4bfbfddfPgPiik:l:nnpiqissuMvMxxz1{1}}kk݉݋OO33  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2IW2lIW2ӘW2ӘW2 W2 W2ɱW2@ɱW2gW2gW2W2W2X2@X2X2 X2X2X29\X29\X2bX2 bX2wlX2 wlX2oX2d oX2X2 X2X2 X2X2H X2CX2hCX2Y2DY2 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_><|  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d_>|[:xW6  t P  S  2 p   @ p O.,l bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)8*844>?IISS]x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$l l""$T%T'')<*<,,.$/$113 4 6688:h;h==?N@NBBD2E2GGIJLLNNPlQlSSUPVPXXZ4[4]]_`bKbKddf/g/iiklnnp:q:ssuvxxz{}t}tXXʄʆ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2SX2 SX2X28 X2X2XX2kY24kY2\Y2\Y2Y2,Y2Z2Z2)[2)[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U_> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U_> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U_>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1~zz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18F ~  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  WW)C*C44>/?/IIST^^hhrsss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccWW K K  ??33''  "#%%'')q*q,,.]/]113I4I668595;;=!>!@@B C EEGGIoJoLLNWOWQQS?T?VVX'Y'[[]^``bbdkekggiSjSlln;o;qqs#t#vvx y {{}}ggۂۄOOÇÉ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2UW2TUW2ߘW2ߘW2W2W2W2(W2[W2[W2'W2'W2X2(X2X2X2X2X2-\X2-\X2bX2 bX2lX2 lX2oX2L oX2X2 X2{X2 {X2X20 X2X2PX2Y2,Y2]Y2]Y2#Y2$#Y2Z2Z25[25[2}P\2}P\2M"]2M"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _wB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _wB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _wB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1. n  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hC y : } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  XX)J*J44>7@@B)C)EEGHJJL M OOQQSpTpVVX^Y^[[]L^L``b8c8eeg"h"jjl m ooqqsetevvxKyK{{}1~1ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2gGW2gGW2W2LW2OW2TOW2٘W2٘W2W2W29W2(9W2W2W2!W2!W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2L oX2X2 X2X2 X2_X28 _X2X2XX2Y24Y2\Y2\Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V_> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V_> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V_>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1J b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T  b ) E } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  SS);*;44>#?#IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33## { {""$g%g'')O*O,,.7/711346689;{;{==?c@cBBDKEKGGI3J3LLNOQQSSUmVmXXZQ[Q]]_5`5bbdeggiikolonnpSqSssu7v7xxz{}}qqUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2UW2TUW2ߘW2ߘW2W2W2W2(W2[W2[W2'W2'W2X2(X2X2X2X2X2-\X2-\X2bX2 bX2lX2 lX2oX2L oX2X2 X2}X2 }X2X20 X2X2PX2% Y2,% Y2^Y2^Y2 Y2$ Y2Z2Z2[2[2 P\2 P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _wC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _wC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _wC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T / e  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>.?.IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIVJVLLN@O@QQS*T*VVXY[[]]_s`sbbdYeYggi?j?lln#o#qqstvvvvxxzX{X}}::ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W2?W2(?W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2_X2 _X2X28 X2X2XX2Y24Y2\Y2\Y2Y2,Y2Z2Z2[2[2O\2O\2!]2!]2M4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @a@ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @a@ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @a@  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1o} b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R$v 0p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10``  KK)5*544>?IIST^v^vhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1``LL 8 8  $$ppZ Z""$D%D'').*.,,./11346w6w88:a;a==?I@IBBD/E/GGIJLLNNPnQnSSUTVTXXZ:[:]]_ ` bbdegygyiik_l_nnpEqEssu+v+xxz{}}jj݄݆PP‰‹44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2W2(W2UW2UW23W23W2X2(X2X2X2X2X2'\X2'\X2cX2 cX2lX2 lX2oX2L oX2X2 X25X2 5X2X20 X2X2PX2Y2,Y2K^Y2K^Y2)Y2$)Y2Z2Z2;[2;[2P\2P\2_"]2_"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ M bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ M bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ M  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Yy b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@& r  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>6?6IIS"T"^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGIvJvLLNbObQQSNTNVVX:Y:[[]&^&``bceeggitjtlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W29W2(9W2W2W2'W2'W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2_X28 _X2X2XX2'Y24'Y2I]Y2I]Y2Y2,Y2Z2Z2[2[2uO\2uO\2E!]2E!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @`? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @`? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @`?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1`\\4CcG b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1   bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 H  l >  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  KK)1*144>?IISS]h^hhhrJsJ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&nnT T""$:%:'') * ,,./1y1y335_6_88:E;E==?)@)BBD E G|G|IIK^L^NNP@Q@SSU"V"XXZ[]u]u__aWbWddf9g9iiklnnpprnsnuuwPxPzz|2}2ff֋֍FF b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2W2(W2UW2UW23W23W2X2(X2X2X2X2X2'\X2'\X2cX2 cX2lX2 lX2oX2L oX2X2 X2SX2 SX2X20 X2X2PX2YY2,YY2^Y2^Y2Y2$Y2Z2Z2#[2#[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _N bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _N bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _N  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations16  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 X > $ f } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>(?(IIST^^hhrlsl} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI8J8LLN O QQSTV|V|XXZd[d]]_L`Lbbd4e4ggijllnoqxqxssu`v`xxzH{H}}00tt\\ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2mW2TmW2W2W2-W2-W2EW2(EW2W2W2?W2?W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X28 X2%X2X%X2Y24Y2]Y2]Y2Y2,Y2Z2Z2)[2)[2#P\2#P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IwV bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IwV bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IwV  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1d b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations16v bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X5 o 6 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  PP):*:44>%?%IIS T ^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccSS C C  33##o o""$Y%Y'')C*C,,.-/-11346689;v;v==?`@`BBDJEJGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--uu__ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2[W2T[W2W2W2W2W2W2(W2IW2IW2-W2-W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2)X2 )X2X20 X2X2PX2Y2,Y2_Y2_Y2#Y2$#Y2Z2Z25[25[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _wA bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _wA bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _wA  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ll b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1: z  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 z k  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  TT)B*B44>0?0IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffTT B B  00  q q""$_%_'')M*M,,.;/;113)4)6689;;=>@|@|BBDjEjGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd_e_ggiIjIlln3o3qqstvvxy{|{|}}ffۂۄPPŇʼn::$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2mW2TmW2W2W2-W2-W2?W2(?W2W2W2?W2?W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X28 X2aX2XaX2!Y24!Y2!^Y2!^Y2 Y2, Y2Z2Z2[2[2O\2O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JvU bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JvU bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JvU  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1YUU-S b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1L )  c  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  NN)8*844>"?"IIST^y^yhhr_s_} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccOO ; ;  ''ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?N@NBBD8E8GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆOO‰‹55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2[W2T[W2W2W2W2W2W2(W2OW2OW2-W2-W2X2(X2X2X2X2X2!\X2!\X2bX2 bX2lX2 lX2oX2L oX2X2 X2MX2 MX2X20 X2X2PX2O Y2,O Y2_Y2_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _wB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _wB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _wB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 + bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\ M > /7w } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ);*;44>%?%IIS T ^~^~hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI)J)LLNOQQSSUhVhXXZN[N]]_4`4bbdeggijlslsnnpYqYssu?v?xxz%{%}} ~~dd׉׋JJ00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2 X28  X2OX2XOX2Y24Y2K^Y2K^Y2#Y2,#Y2Z2Z25[25[2MP\2MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @`? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @`? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @`?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Yy b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10 h  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>/?/IIST^^hissss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGIWJWLLNEOEQQS3T3VVX!Y![[]^``bbdtetggibjbllnKoKqqs/t/vvxy{{}}iiۂۄMM11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2[W2T[W2W2W2W2W2W2(W2CW2CW2-W2-W2X2(X2X2X2wX2wX2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2 X20  X2OX2POX2Y2,Y2_Y2_Y2#Y2$#Y2Z2Z25[25[2;P\2;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m9  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1| bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pl h d L } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)F*F44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj^^ R R  FF::**  "#%x%x'')d*d,,.P/P113<4<668(9(;;=>@@BBDpEpGGIXJXLLN@O@QQS(T(VVXY[[]]_l`lbbdTeTggi#?#IIS T ^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccMM 7 7  !!  jjT T""$>%>'')(*(,,./11335q6q88:[;[==?E@EBBD/E/GGIJLLNOQzQzSSUdVdXXZN[N]]_8`8bbd"e"ggi j llnnpeqessuIvIxxz-{-}}ggىًKK// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2[W2T[W2W2W2W2W2W2(W2IW2IW2-W2-W2X2(X2X2X2}X2}X2\X2\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2aX2PaX2Y2,Y2`Y2`Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2;4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 @` b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ` \ XTH } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)<*<44>"?"IIST^x^xhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((p p""$X%X'')@*@,,.(/(11346688:l;l==?R@RBBD6E6GGIJLLNNPpQpSSUTVTXXZ8[8]]_`bbdegrgriikVlVnnp:q:ssuvxxz{}t}tXXʄʆ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2KW2(KW2W2W23W23W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2eX2 eX2X28 X2X2XX2Y24Y2_Y2_Y2)Y2,)Y2Z2Z2;[2;[2_P\2_P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @`? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @`? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @`?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nnz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1{n a  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  [[)G*G44>3?3IIST^^hiswsw} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77++  "#%%'')u*u,,.a/a113M4M668999;;=%>%@@BCEEGGIsJsLLN[O[QQSCTCVVX+Y+[[]^``bbdoeoggiWjWlln?o?qqs't'vvxy{{}}kk߂߄SSLJlj;;## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2aW2TaW2W2W2!W2!W2W2(W2=W2=W23W23W2X2(X2X2X2qX2qX2\X2\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2[X2P[X2Y2,Y2`Y2`Y25Y2$5Y2 Z2 Z2G[2G[2P\2P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb. b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _y> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb. b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _y> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1vb.  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _y>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .Nw b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HO V ] R } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJLMOOQQStTtVVXbYb[[]P^P``b'?'IIS T ^}^}hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''  ""$k%k'')S*S,,.;/;113#4#668 9 ;;==?g@gBBDOEOGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2aW2TaW2W2W2!W2!W2W2(W2CW2CW23W23W2X2(X2X2X2wX2wX2\X2\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X21X2P1X2Y2,Y2oaY2oaY2#Y2$#Y2Z2Z25[25[2P\2P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _x> bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _x> bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _x>  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 = b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 ]  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d k ry } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNDODQQS.T.VVXY[[]^`w`wbbd]e]ggiCjClln'o'qqs t vzvzxxz\{\}}>> ttVV b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2gW2TgW2W2W2'W2'W2QW2(QW2W2W29W29W2X2(X2X2X2#X2#X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2}X2 }X2X28 X2X2XX2Y24Y2_Y2_Y2Y2,Y2Z2Z2)[2)[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @`? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @`? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @`?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1GC b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1t  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`^ \ Z H } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  OO)9*944>!?!IIST^z^zhhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnTTƉƋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2eW2LeW2gW2TgW2W2W2'W2'W2W2(W27W27W29W29W2X2(X2X2X2kX2kX2 \X2 \X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2_X20 _X2X2PX2Y2,Y2oaY2oaY2AY2$AY2Z2Z2S[2S[2P\2P\2w"]2w"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1kW#  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 !Aa. b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 2 D V  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)J*J44>:?:IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$r%r'')b*b,,.R/R113B4B668292;;=">"@@BCEEGHJzJzLLNfOfQQSRTRVVX>Y>[[]*^*``bceeghjxjxllndodqqsPtPvvx?IISS]l^lhhrNsN} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **rrX X""$>%>'')$*$,,. / 1}1}335c6c88:I;I==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6jjڋڍJJ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2gW2TgW2W2W2'W2'W2W2(W2=W2=W29W29W2X2(X2X2X2qX2qX2\X2\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X27X2P7X2 Y2, Y2AbY2AbY2)Y2$)Y2Z2Z2;[2;[2/P\2/P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _? bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _? bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m['  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _?  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 +K b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1  bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 h z  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)@*@44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$R%R'')>*>,,.*/*11346689;x;x==?d@dBBDPEPGGI8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88%%  ""$n%n'')X*X,,.B/B113,4,6689;;=>@u@uBBD_E_GGIHJHLLN0O0QQSTVVXY[t[t]]_\`\bbdDeDggi,j,llnoqqssupvpxxzX{X}}@@((pp b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W21W21W2W2W2eX2peX2\X2\X2bX2bX2cX2`cX2X2X23X23X2[NY2L [NY2AY2 AY2Z2p Z2S[2t S[2wP\2T wP\2G"]2G"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_K b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_K b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_K b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  aa)O*O44>;?;IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssaa O O  ==++ ~ ~""$l%l'')Z*Z,,.H/H113646668$9$;;=>@@BCEwEwGGIcJcLLNMOMQQS7T7VVX!Y![[] ^ ``bbdjejggiTjTlln>o>qqs(t(vvxy{{}}qq[[ЇЉEE// b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W25W2gW2$gW2iW2iW2W2W29W2$9W2;X2;X2[X2[X2 `X2$ `X2ŢX2ŢX2%X2%X23X2 3X2[NY2| [NY2/Y2 /Y2Z2 Z2A[2 A[2O\2 O\2!]2!]2w4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14u b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14u b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14u b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  aa)K*K44>3?3IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvbb N N  ::&&p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI3J3LLNOQQSSUrVrXXZX[X]]_>`>bbd"e"ggijlxlxnnp\q\ssu@v@xxz${$}}zz^^щыDD** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2ťX2`ťX2%X2%X23X23X2[NY2L [NY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1_M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ``)J*J44>2?2IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$r r""$]%]'')G*G,,.1/111346689;z;z==?d@dBBDNENGGI6J6LLNOQQSTVuVuXXZ[[[]]_A`Abbd'e'ggi j llnnpfqfssuLvLxxz2{2}}qqWWʎʐ== b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2gW2gW2cW2cW2W2W29W29W25X2p5X2[X2[X2 `X2 `X2X2`X2uX2uX23X23X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1#C  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>?>IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ttbb P P  >>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIfJfLLNTOTQQSBTBVVX0Y0[[]^``b c eeggiqjqllnZoZqqs>t>vvx"y"{{}~xx\\·Ή@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2 X2` X2iX2iX23X23X2UNY2L UNY25Y2 5Y2 Z2p  Z2G[2t G[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hT  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hT  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hT  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  cc)O*O44>7?7IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ssgg [ [  OOCC33  " # %%'')m*m,,.Y/Y113E4E668191;;=>@@BCEwEwGGI_J_LLNGOGQQS/T/VVXY[[]]_s`sbbd[e[ggiCjClln+o+qqstvvxxzo{o}}WW˂˄??$$ }} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2gW2gW2cW2cW2W2W29W29W25X2p5X2[X2[X2 `X2 `X2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2)Y2 )Y2Z2p Z2;[2t ;[2 P\2T  P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1r b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  \\)F*F44>0?0IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr\\ F F  00yyc c""$M%M'')7*7,,.!/!113 4 6688:j;j==?T@TBBD>E>GGI(J(LLNOQQSSUqVqXXZ[[[]]_E`Ebbd/e/ggijllnoqrqrssuVvVxxz:{:}}ttXXʎʐ<< b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hV" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hV" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1hV" b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss  __)G*G44>+?+IIST^^hhrcsc} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1sscc S S  CC33 { {""$c%c'')K*K,,.3/311346689;w;w==?[@[BBD?E?GGI#J#LLNOQyQySSU]V]XXZA[A]]_%`%bbd e g{g{iik_l_nnpCqCssu'v'xxz { }}}}aaӄӆCC%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2mW2mW2]W2]W2W2W2?W2?W2/X2p/X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2qP\2T qP\2A"]2A"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  hh)T*T44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tthh \ \  PPDD88,,  " # %%' ( **,,.n/n113Z4Z668F9F;;=2>2@@BCEEG H J~J~LLNfOfQQSNTNVVX6Y6[[]^``bcezezggibjbllnJoJqqs2t2vvxy{{}~vv^^҇҉FF.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2׎W2׎W2W2W2=W2=W2W2W2qX2pqX2\X2\X2{bX2{bX2X2`X2oX2oX2-X2-X2UNY2L UNY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q]) b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q]) b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q]) b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!A  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  aa)S*S44>C?CIIS/T/^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ooaa S S  EE77))  " # %%'')x*x,,.j/j113\4\668N9N;;=@>@@@B2C2EEG$H$JJLMOOQRTwTwVVXeYe[[]S^S``b?c?eeg)h)jjlmooqqsltlvvxRyR{{}8~8ttXXʑʓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2mW2mW2WW2WW2W2W2?W2?W2)X2p)X2[X2[X2`X2`X2X2`X2WX2WX2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1'  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)J*J44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%v%v'')^*^,,.F/F113.4.6689;;==?r@rBBDZEZGGI@J@LLN$O$QQSTVzVzXXZ^[^]]_B`Bbbd&e&ggi j l|l|nnp`q`ssuDvDxxz({(}} ~~bbԉԋFF** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2׎W2׎W2W2W2=W2=W2W2W2qX2pqX2\X2\X2{bX2{bX2X2`X2X2X2-X2-X2UNY2L UNY2)Y2 )Y2Z2p Z2;[2t ;[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q_+ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo  ]])K*K44>7?7IIST^^hisqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1oo]] K K  99'' z z""$h%h'')V*V,,.D/D113242668 9 ;;=>@@BBDsEsGGI_J_LLNIOIQQS3T3VVXY[[]^`|`|bbdbebggiHjHlln,o,qqstvvxxza{a}}CC%%wwYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2[W2[W2WW2WW2W2W2-W2-W2)X2p)X2[X2[X2_X2_X2ťX2`ťX2%X2%X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  XX)B*B44>(?(IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11}}g g""$Q%Q'');*;,,.%/%11346688:n;n==?T@TBBD:E:GGI J LLNOQyQySSU_V_XXZE[E]]_+`+bbdeggiikjljnnpPqPssu6v6xxz{}}uuYYˉˋ==!! b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2;Y2 ;Y2Z2p Z2M[2t M[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zf2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zf2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zf2 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  [[)K*K44>9?9IIS%T%^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kk[[ K K  ;;++   ""$s%s'')c*c,,.S/S113C4C668393;;=#>#@@BCEEGHJyJyLLNeOeQQSQTQVVX=Y=[[])^)``bceeghjwjwllncocqqsOtOvvx;y;{{}'~'uuaaבד b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2[W2[W2QW2QW2W2W2-W2-W2#X2p#X2[X2[X2_X2_X2WX2`WX2X2X2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2[2t [2O\2T O\2W!]2W!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1l b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1n  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  VV)<*<44>?IIST^q^qhhrQsQ} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11yy_ _""$E%E'')+*+,,./11335j6j88:P;P==?2@2BBDEGGIIKgLgNNPIQISSU+V+XXZ [ ]~]~__a`b`ddfBgBiik$l$nnpqswswuuwYxYzz|;};mm݋ݍMM b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2CW2CW2W2W2wX2pwX2\X2\X2bX2bX2X2`X2EX2EX2'X2'X2ONY2L ONY2#Y2 #Y2Z2p Z25[2t 5[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zh4 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zh4 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zh4 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk  WW)C*C44>-?-IIST^^hhrqsq} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1kkWW C C  //}}i i""$U%U'')A*A,,.-/-11346689;{;{==?g@gBBDSESGGI=J=LLN%O%QQS T VVXXZi[i]]_Q`Qbbd9e9ggi!j!lln o q}q}ssuevexxzM{M}}55yyaa b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2gW2gW2QW2QW2W2W29W29W2#X2p#X2[X2[X2 `X2 `X2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2/[2t /[2)P\2T )P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  YY)C*C44>,?,IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, x x""$b%b'')L*L,,.6/6113 4 668 9 ;;==?i@iBBDSESGGI/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggUU C C  11  r r""$`%`'')N*N,,.@}@}BBDkEkGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2gW2gW2KW2KW2W2W29W29W2X2pX2[X2[X2 `X2 `X2cX2`cX2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1i b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1y  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  UU)?*?44>'?'IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjVV B B  ..zzd d""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiiklllnnpPqPssu4v4xxz{}}nnRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2юW2юW2W2W2IW2IW2W2W2}X2p}X2\X2\X2ubX2ubX2X2`X2X2X2'X2'X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1x} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1x} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1x } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  TT)>*>44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,zzf f""$Q%Q'');*;,,.%/%11346688:n;n==?X@XBBDBEBGGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2UW2UW2KW2KW2W2W2'W2'W2X2pX2[X2[X2_X2_X2QX2`QX2X2X2!X2!X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1m  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>2?2IIS T ^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIZJZLLNHOHQQS6T6VVX$Y$[[]^``bcewewggiejellnNoNqqs2t2vvxy{{}}llނބPP‡‰44 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2}W2 }W2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2ťX2`ťX2%X2%X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xD b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  WW)C*C44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1gg[[ O O  CC77''  ""$u%u'')a*a,,.M/M113949668%9%;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33qq b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2UW2UW2EW2EW2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>$?$IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ{Q{SSUeVeXXZO[O]]_9`9bbd#e#ggi j llnnpfqfssuJvJxxz.{.}}hhډڋLL00 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2}W2 }W2W2W2W2W2OW2OW2W2W2X2pX2!\X2!\X2bX2bX2WX2`WX2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2[2t [2O\2T O\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zF b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg  SS);*;44>?IIST^u^uhhrWsW} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ggWW G G  77''o o""$W%W'')?*?,,.'/'11346688:k;k==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2aW2aW2?W2?W2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2/[2t /[2SP\2T SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>2?2IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGIrJrLLNZOZQQSBTBVVX*Y*[[]^``bbdnenggiVjVlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2ŎW2ŎW2W2W2OW2OW2W2W2X2pX2!\X2!\X2ibX2ibX2?X2`?X2X2X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2S"]2S"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1M b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1_  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  UU)G*G44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccUU G G  99++  "#%z%z'')l*l,,.^/^113P4P668B9B;;=4>4@@B&C&EEGHJJLMO}O}QQSkTkVVXYYY[[]G^G``b3c3eeghjjlmooqqsjtjvvxPyP{{}6~6rrVVȑȓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2[W2[W2?W2?W2W2W2-W2-W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2CNY2L CNY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1@`  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)>*>44>$?$IIST^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI4J4LLNOQQSSUnVnXXZR[R]]_6`6bbdeggiikplpnnpXqXssu>"" b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2ŎW2ŎW2W2W2UW2UW2W2W2X2pX2'\X2'\X2ibX2ibX2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc  QQ)?*?44>+?+IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ccQQ ? ?  --  n n""$\%\'')J*J,,.8/8113&4&6689;;=>@y@yBBDgEgGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdVeVggi?IIST^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss] ]""$G%G'')1*1,,./11346z6z88:d;d==?J@JBBD0E0GGIJLLNNPoQoSSUUVUXXZ;[;]]_!`!bbdeggiiklllnnpRqRssu8v8xxz{}}ww[[͉͋??## b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2UW2UW2W2W2X2pX2'\X2'\X2bX2bX2X2`X2cX2cX2X2X2=NY2L =NY25Y2 5Y2]Z2p ]Z2;[2t ;[2P\2T P\2 ]2 ]293]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1V b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1JF  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  OO)I*I44>7?7IIS#T#^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__OO ? ?  //w w""$q%q'')a*a,,.Q/Q113A4A668191;;=!>!@@BCEEGHJwJwLLNcOcQQSOTOVVX;Y;[[]'^'``bceeggiujullnmomqqsYtYvvxEyE{{}1~1 kk b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2gGW2gGW2W2 W2OW2OW29W29W2W2W2!W2!W2 X2p X2[X2[X2_X2_X2ߣX2`ߣX2?X2?X2X2X2=NY2T =NY2Y2 Y2!Z2x !Z2[2| [2{O\2\ {O\2q]2q]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1`  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1V!  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  NN)4*444>?IIST^u^uhhrUsU} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?*@*BBD E G}G}IIK_L_NNPAQASSU#V#XXZ[]v]v__adbdddfFgFiik(l(nnp q s{s{uuw]x]zz|?}?!!qqQQ b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2UW2UW2W2W2X2pX2'\X2'\X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y2EZ2p EZ2#[2t #[2#P\2T #P\2 ]2 ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1X b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10__  KK)?*?44>)?)IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1__KK 7 7  ##qq] ]""$Q%Q'')=*=,,.)/)11346689;w;w==?c@cBBDOEOGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqqssumvmxxzU{U}}==%% ii b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations12W2GW2GW2W2|W2W2W2QW2QW2ۙW2|ۙW2W2HW2W2W2W2PW2W2XW2#W2#W2X2PX2X2X2X2X2[[X2$ [[X2`X2, `X2cX2 cX2mX2$ mX2pX2 pX2X2| X2X2XX2 X2 X2MX2MX2Y2Y2=1?1IIS%T%^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11SS? ?""$y%y'')e*e,,.Q/Q113=4=668)9);;=>@@BBDqEqGGIYJYLLNAOAQQS)T)VVXY[[]]_m`mbbdaeaggiIjIlln1o1qqstvvxy{u{u}}]]тфEE,,# b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2OW2lOW2٘W2٘W2W2W2ϱW2@ϱW2mW2mW2!W2!W2X2@X2X2 X2X2X2?\X2?\X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2X2 X2X2H X2=X2h=X2Y2DY2O ? IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%// ""$G%G'')1*1,,./11346z6z88:d;d==?L@LBBD2E2GGIJLLNNPqQqSSUWVWXXZ=[=]]_#`#bbdeggiiknlnnnpTqTssu:v:xxz { }}yy__щыCC''  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations12W2GW2GW2W2|W2W2W2QW2QW2ۙW2|ۙW2W2HW2W2W2W2PW2W2XW2#W2#W2X2PX2X2X2X2X2U[X2$ U[X2`X2, `X2cX2 cX2mX2$ mX2pX2 pX2X2| X2X2TX2gX2gX2X2X2Y2Y2;Y2T;Y2EaY2xEaY2ClY2ClY2Y20Y2Z2Z2[2[2[2[2O\2dO\2!]2!]254]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 bWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1  bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1O  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 okz2R b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1  J   b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d%n7          BA bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  )=*=44>#?#IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!!<<$ $""$Y%Y'')A*A,,.)/)11346688:m;m==?S@SBBD7E7GGIJLLNNPqQqSSUUVUXXZ9[9]]_`bbd e ggiikclcnnpGqGssu+v+xxz{}}eeׄ׆II++ # b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2HW2HW2W2dW2OW2lOW2٘W2٘W2W2W2ձW2@ձW2sW2sW2!W2!W2X2@X2X2 X2X2X2E\X2E\X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2X2 X2X2H X2SX2hSX2 Y2D Y2k>Y2k>Y2aY2aY2lY2 lY2Y2Y2Z2pZ2[2t[2[2T[2uO\2uO\2E!]2$E!]23]2(} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1x b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a^B bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1x b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a^B bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1x  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a^B  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1  U  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1d (d2+<rFPX  `  h  p  x * Z$t bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa    )4*444>?IIST^w^whhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%,, ""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__afbfddfHgHiik*l*nnp q s}s}uuw_x_zz|A}A##uuUUŐŒ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2gGW2gGW2W2dW2OW2lOW2٘W2٘W2W2W29W2@9W2W2W2!W2!W2X2@X2X2 X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2X2 X2YX2P YX2X2pX2Y2LY2=7?7IISS]^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaSS E E  77))   ""$x%x'')j*j,,.\/\113N4N668@9@;;=2>2@@B$C$EEGHJJLMO}O}QQSkTkVVXYYY[[]G^G``bbddedggiNjNllnoqqsjtjvvxPyP{{}6~6ttXXʑʓ  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2OW2lOW2٘W2٘W2W2W2ɱW2@ɱW2gW2gW2!W2!W2X2@X2X2 X2X2X29\X29\X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2ѥX2 ѥX21X2H 1X2sX2hsX2Y2DY2=&?&IISS]^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddTT D D  44$$p p""$Z%Z'')D*D,,../.11346689;w;w==?a@aBBDKEKGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jb|b|ddfdgdiikLlLnoqxqxssu`v`xxzH{H}}00xxbb  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2gGW2gGW2W2dW2OW2lOW2٘W2٘W2W2W29W2@9W2W2W2!W2!W2X2@X2X2 X2 X2 X2[X2[X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2X2 X2X2D X2WX2dWX2kY2<kY2'=Y2'=Y2^Y2^Y2iY2iY2Y2Y2Z2hZ2#[2l#[2[2L[2 P\2 P\2!]2!]24]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1a   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 fbh b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1a j  K  b-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5z*z*breasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1 bJ7Moperation.5exchange :z2breasonOperatorInputsvalue9invocations1daB# ehF0'xiJ+   m  N  /  qRT bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IISS]^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdeggiikjljn(o(qqs t v{v{xxz]{]}}??!!uuWW  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1x2W2GW2GW2W2dW2OW2lOW2٘W2٘W2W2W2ɱW2@ɱW2gW2gW2!W2!W2X2@X2X2 X2X2X29\X29\X2bX2 bX2}lX2 }lX2oX2d oX2X2 X2X2 X2X2H X2/X2h/X2Y2DY2CiJ+   m  N  /   qRT bWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  MM)7*744>!?!IISS]|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&rr\ \""$F%F'')0*0,,./11346y6y88:c;c==?M@MBBD7E7GGI!J!LLNOQzQzSSU`V`XXZF[F]]_,`,b]b]ddfAgAiik%l%nnpNqNssu2v2xxz{}}llބކRRʼnŋ88  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W2?W2(?W2W2W2'W2'W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2SX28 SX2X2XX2aX24aX2TY2TY2#Y2,#Y2Z2Z25[25[2P\2P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UaB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UaB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UaB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1; b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1I[H } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>1?1IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$i%i'')Y*Y,,.I/I113949668)9);;=>@@B C EEGGIqJqLLN]O]QQSITIVVX5Y5[[]!^!``b c eeggiojolln[o[qqsGtGvvx3y3{{}~ mmYYϑϓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2[W2T[W2W2W2W2W2ñW2(ñW2aW2aW2-W2-W2X2(X2X2X2X2X23\X23\X2bX2 bX2lX2 lX2oX2L oX2X2 X2GX2 GX2X20 X2X2PX2aX2,aX2G\Y2G\Y2Y2$Y2Z2Z2)[2)[2AP\2AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 auF bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 auF bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1^  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 auF  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]:  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  RR)@*@44>0?0IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddRR @ @  ..  o o""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGIXJXLLNFOFQQS4T4VVX"Y"[[]^``bbdueuggicjcllnLoLqqs0t0vvxy{{}}jj܂܄NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2mGW2mGW2 W2L W2UW2TUW2ߘW2ߘW2W2W29W2(9W2W2W2'W2'W2X2(X2X2X2 X2 X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2GX2, GX2X2LX2X2$X2WUY2WUY2 Y2 Y2Z2Z2[2[2P\2P\2!]2!]24]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 VaB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 VaB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b   b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 VaB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 [WW * b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$m6 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM)9*944>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$K%K'')7*7,,.#/#11346688:q;q==?]@]BBDIEIGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--qqYY b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2[W2T[W2W2W2W2W2ñW2(ñW2aW2aW2-W2-W2X2(X2X2X2X2X23\X23\X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2)X2P)X2Y2,Y2\Y2\Y2Y2$Y2Z2Z2[2[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 auG bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 auG bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 auG  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 l b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tQ. ] } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  LL)6*644>"?"IIS T ^~^~hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbLL 6 6      iiS S""$=%='')'*',,./11335p6p88:Z;Z==?D@DBBD.E.GGIJLLNOQyQySSUcVcXXZM[M]]_7`7bbd!e!ggi j llnnpdqdssuHvHxxz,{,}}ff؉؋JJ.. b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2[W2T[W2W2W2W2W2EW2(EW2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2X2 X2SX28 SX2X2XX2qX24qX2-UY2-UY2Y2,Y2Z2Z2/[2/[2SP\2SP\2#"]2#"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @cD bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @cD bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @cD  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations16 b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 L bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1V*~ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hhrtst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLN=O=QQS'T'VVXY[[]]_p`pbbdZeZggiDjDlln.o.qqstvvxy{w{w}}aaւքKK55 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2gW2TgW2W2W2'W2'W2W2(W2UW2UW29W29W2X2(X2X2X2X2X2'\X2'\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2GX2 GX2X20 X2X2PX2aX2,aX2\Y2\Y2Y2$Y2Z2Z2/[2/[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 aQ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 aQ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 aQ  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hP8p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  \\)H*H44>4?4IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hh\\ P P  DD88,,   "#%%'(*v*v,,.b/b113N4N668:9:;;=&>&@@BCEEGGItJtLLN\O\QQSDTDVVX,Y,[[]^``bbdpepggiXjXlln@o@qqs(t(vvxy{{}}llTTȇȉ<<$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2[W2T[W2W2W2W2W2?W2(?W2W2W2-W2-W2X2(X2X2X2X2X2[X2[X2bX2 bX2lX2 lX2oX2L oX2X2 X2]X2 ]X2X2, X2X2LX2 Y2$ Y2UY2UY2 Y2 Y2Z2Z2[2[2O\2O\2!]2!]2e4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1c  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 PLLz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R&zN6v } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  NN)8*844>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbNN : :  &&tt` `""$K%K'')5*5,,./113 4 6~6~88:h;h==?R@RBBD*>44>&?&IIS T ^|^|hhr`s`} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$j%j'')R*R,,.:/:113"4"668 9 ;~;~==?f@fBBDNENGGI6J6LLNOQQSSUpVpXXZT[T]]_8`8bbdeggijlrlrnnpVqVssu:v:xxz{}}ttXXʉʋ<<  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2sW2TsW2W2W23W23W2EW2(EW2W2W2EW2EW2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2qX28 qX2X2XX2Y24Y2KUY2KUY2Y2,Y2Z2Z2/[2/[25P\25P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IyZ bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IyZ bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IyZ  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1eaazS b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1%e bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X0 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)A*A44>+?+IIST^^hhroso} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeYY M M  AA55%%  ""$s%s'')_*_,,.K/K113747668#9#;;=>@@BBDkEkGGISJSLLN;O;QQS#T#VVX Y [[]]_g`gbbdOeOggi7j7llnoqqstv{v{xxzc{c}}KK33ss b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2W2(W2OW2OW23W23W2X2(X2X2X2X2X2!\X2!\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2 X20  X2KX2PKX2gX2,gX2]Y2]Y2 Y2$ Y2Z2Z2[2[2O\2O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 auE bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 auE bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 auE  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1sfY ` } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^{^{hhrasa} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?N@NBBD4E4GGIJLLNOQsQsSSUYVYXXZ?[?]]_%`%bbd e g~g~iikdldnnpJqJssu0v0xxz{}}ooUUljNj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2sGW2sGW2W2LW2sW2TsW2W2W23W23W2EW2(EW2W2W2EW2EW2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X2, X2]X2L]X2+Y2$+Y2UY2UY2Y2Y2Z2Z2[2[2O\2O\2!]2!]254]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JxY bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JxY bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JxY  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 EAA b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`8hX } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)9*944>?IIST^u^uhhrYsY} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%m m""$U%U'')=*=,,.%/%113 4 6688:i;i==?O@OBBD3E3GGIJLLNNPmQmSSUQVQXXZ5[5]]_`bbddfogoiikSlSnnp7q7ssuvxxzz|q}qUUDŽdž99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2aW2TaW2W2W2!W2!W2W2(W2UW2UW23W23W2X2(X2X2X2X2X2'\X2'\X2cX2 cX2lX2 lX2oX2L oX2X2 X2դX2 դX25X20 5X2wX2PwX2Y2,Y2c^Y2c^Y2Y2$Y2Z2Z2 [2 [2{O\2{O\2K!]2K!]23]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 auF bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 auF bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1L  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 auF  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Zz b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1L?2% } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  NN)4*444>?IISS]k^khhrMsM} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))qqW W""$=%='')#*#,,. / 1|1|335b6b88:H;H==?,@,BBDEGGIIKaLaNNPCQCSSU%V%XXZ[]x]x__aZbZddf;?;IIS'T'^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeWW I I  ;;--  "#%|%|'')n*n,,.`/`113R4R668D9D;;=6>6@@B(C(EEGHJJL M OOQQSoToVVX]Y][[]K^K``b7c7eeg!h!jjl m o~o~qqsdtdvvxJyJ{{}0~0nnRRđē b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2aW2TaW2W2W2!W2!W2W2(W2IW2IW23W23W2X2(X2X2X2}X2}X2\X2\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2?X2P?X2mX2,mX2c^Y2c^Y2Y2$Y2Z2Z2/[2/[2)P\2)P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 aC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q= b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 aC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q=  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 aC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ,L b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~|z } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  UU)?*?44>*?*IIST^^hhrpsp} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhXX H H  88((t t""$^%^'')H*H,,.2/211346689;{;{==?e@eBBDOEOGGI:J:LLN"O"QQS T V~V~XXZf[f]]_N`Nbbd6e6ggijllnoqzqzssubvbxxzJ{J}}22zzdd b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2yGW2yGW2W2LW2aW2TaW2W2W2!W2!W2KW2(KW2W2W23W23W2X2(X2X2X2X2X2[X2[X2cX2 cX2lX2 lX2oX2L oX2X2 X2wX2 wX2X2, X2X2LX2X2$X2MVY2MVY2Y2Y2Z2Z2/[2/[2P\2P\2!]2!]24]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :66~ b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 oR5 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbdZeZggi@j@lln$o$qqstvwvwxxzY{Y}};;qqSS b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2}W2L}W2aW2TaW2W2W2!W2!W2W2(W2OW2OW23W23W2X2(X2X2X2X2X2!\X2!\X2cX2 cX2lX2 lX2oX2L oX2X2 X2X2 X2X20 X2aX2PaX2+Y2,+Y2_Y2_Y2Y2$Y2Z2Z2#[2#[2O\2O\2!]2!]2_4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 aC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 aC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1uA  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 aC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Qq b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xvtr } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>%?%IIS T ^|^|hhrbsb} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?Q@QBBD;E;GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikjljnnpNqNssu2v2xxz{}}llބކRRʼnŋ88 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2gW2TgW2W2W2'W2'W2QW2(QW2W2W29W29W2X2(X2X2X2#X2#X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2SX28 SX2X2XX2cY24cY2VY2VY25Y2,5Y2 Z2 Z2G[2G[2P\2P\2k"]2k"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1OKK& b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations11 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1n\Jp } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  UU)E*E44>5?5IIS!T!^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%% } }""$m%m'')]*],,.M/M113=4=668-9-;;=>@@B C EEGGIuJuLLNaOaQQSMTMVVX9Y9[[]%^%``bceeggisjslln_o_qqsKtKvvx7y7{{}#~#qq]]ӑӓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2gW2TgW2W2W2'W2'W2W2(W2CW2CW29W29W2X2(X2X2X2wX2wX2\X2\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X2[X2P[X2mX2,mX2_Y2_Y2)Y2$)Y2Z2Z2;[2;[2YP\2YP\2)"]2)"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 awB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 awB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1xf2  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 awB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :Z b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>4?4IIS"T"^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGI\J\LLNJOJQQS8T8VVX&Y&[[]^``bceyeyggigjgllnPoPqqs4t4vvxy{{}}nnRRćĉ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2W2LW2gW2TgW2W2W2'W2'W2QW2(QW2W2W29W29W2X2(X2X2X2#X2#X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2oX2 oX2X2, X2 X2L X2Y2$Y2VY2VY2#Y2#Y2Z2Z25[25[2)P\2)P\2!]2!]24]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1f  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 /++s b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18&^ } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  QQ)=*=44>)?)IIST^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))wwc c""$O%O'');*;,,.'/'11346688:u;u==?a@aBBDMEMGGI9J9LLN!O!QQS T V}V}XXZe[e]]_M`Mbbd5e5ggijllnoqyqyssuavaxxzI{I}}11uu]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2wW2LwW2gW2TgW2W2W2'W2'W2W2(W2IW2IW29W29W2X2(X2X2X2}X2}X2\X2\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X21X2P1X2Y2,Y2_Y2_Y2Y2$Y2Z2Z2)[2)[2O\2O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zj6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 avB bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zj6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 avB bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1zj6  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 avB  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 < b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$-6?I } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  PP):*:44>&?&IIST^^hhrfsf} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffPP : :  $$mmW W""$A%A'')+*+,,./11335t6t88:^;^==?H@HBBD2E2GGIJLLNOQ}Q}SSUgVgXXZQ[Q]]_;`;bbd%e%ggijllnnphqhssuLvLxxz0{0}}jj܉܋NN22 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2gW2TgW2W2W2'W2'W2WW2(WW2W2W29W29W2X2(X2X2X2)X2)X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2#X2 #X2X28 X2X2XX2sX24sX2_VY2_VY2/Y2,/Y2Z2Z2A[2A[2eP\2eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1D@@X b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1yrk } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  SS)A*A44>/?/IIST^^hisxsx} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeSS A A  //  p p""$^%^'')L*L,,.:/:113(4(6689;;=>@{@{BBDiEiGGIWJWLLNAOAQQS+T+VVXY[[]]_t`tbbd^e^ggiHjHlln2o2qqstvvxy{{{{}}eeڂڄOOćĉ99## b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2kW2LkW2gW2TgW2W2W2'W2'W2W2(W2=W2=W29W29W2X2(X2X2X2qX2qX2\X2\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X2QX2PQX2sX2,sX2_Y2_Y25Y2$5Y2 Z2 Z2G[2G[2kP\2kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m[' b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 aC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m[' b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 aC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m['  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 aC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 : b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@ bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(<0p } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  ``)L*L44>8?8IIS T ^^his|s|} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll`` T T  HH<<00$$  "#%%'(*z*z,,.f/f113R4R668>9>;;=*>*@@BCEEGHJxJxLLN`O`QQSHTHVVX0Y0[[]^``bcetetggi\j\llnDoDqqs,t,vvxy{{}}ppXẊ̉@@(( b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2#W2L#W2gW2TgW2W2W2'W2'W2WW2(WW2W2W29W29W2X2(X2X2X2)X2)X2[X2[X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2YX2, YX2X2LX2/X2$/X2WY2WY2Y2Y2Z2Z2/[2/[2O\2O\2!]2!]2w4]2 } bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @bC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @bC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @bC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b)goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $  h& b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b)goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r9_p16_c72_p72-:-z1211Z5ZYz]"]za"abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements648input0Elements648invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d]VO } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  RR)<*<44>&?&IIS T ^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **xxd d""$O%O'')9*9,,.#/#113 4 6688:l;l==?V@VBBD@E@GGI*J*LLNOQQSSUiViXXZO[O]]_5`5bbdeggijltltnnpZqZssu@v@xxz&{&}} ee؉؋KK11 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1l2W2GW2GW2qW2LqW2gW2TgW2W2W2'W2'W2W2(W2=W2=W29W29W2X2(X2X2X2qX2qX2\X2\X2 cX2  cX2lX2 lX2oX2L oX2 X2  X2X2 X2X20 X27X2P7X2IY2,IY2`Y2`Y2Y2$Y2Z2Z2/[2/[2O\2O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o_+ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 aC bwoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o_+ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 aC bwoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 } bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1o_+  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 _ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 aC  bwoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 b-goperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1} bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  , b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 b-goperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1+ bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Pdx D } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  ZZ)B*B44>*?*IIST^^hhrdsd} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jjZZ J J  ::**   ""$n%n'')V*V,,.>/>113&4&6689;;==?j@jBBDRERGGI:J:LLNOQQSTVtVtXXZX[X]]_<`v>xxz"{"}}xx\\Ή΋@@$$ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2AW2AW2gW2$gW2oW2oW2 W2 W29W2$9W2AX2AX2[X2[X2 `X2$ `X2X2X2oX2oX23X2 3X2[NY2| [NY2AY2 AY2Z2 Z2S[2 S[2YP\2 YP\2)"]2)"]24]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14y& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14y& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14y& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz  jj)V*V44>>?>IIS&T&^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1zznn b b  VVJJ::&&  "#%%'')t*t,,.`/`113L4L668898;;=">"@@B C E~E~GGIfJfLLNNONQQS6T6VVXY[[]^`z`zbbdbebggiJjJlln2o2qqstvvxy{v{v}}^^҂҄FF++ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2eW2 eW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2ˢX2`ˢX2+X2+X23X23X2[NY2L [NY2)Y2 )Y2Z2p Z2;[2t ;[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aO b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aO b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aO b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  ]])G*G44>-?-IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rr^^ J J  66""  l l""$V%V'')@*@,,.*/*11346688:s;s==?Y@YBBD?E?GGI%J%LLN O Q~Q~SSUdVdXXZJ[J]]_0`0bbdeggiikolonnpUqUssu;v;xxz!{!}}zz^^ЉЋBB&& b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2;W2;W2gW2$gW2oW2oW2 W2 W29W2$9W2AX2AX2[X2[X2 `X2$ `X2X2X2cX2cX23X2 3X2[NY2| [NY2)Y2 )Y2Z2 Z2;[2 ;[2O\2 O\2!]2!]2Y4]2} bJ7UGoperation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14w" b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7UGoperation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14w" b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7UGoperation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations14w" b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 *  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx  dd)L*L44>0?0IIST^^hhrhsh} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1xxhh X X  HH88$$   ""$h%h'')P*P,,.8/8113 4 6689;|;|==?`@`BBDDEDGGI(J(LLN O Q~Q~SSUbVbXXZF[F]]_*`*bbdeggiikdldnnpHqHssu,v,xxz{}}ff؄؆HH**  b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W27W27W2W2W2kX2pkX2 \X2 \X2bX2bX2ˢX2`ˢX2+X2+X23X23X2[NY2L [NY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2i!]2i!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aQ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aQ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1aQ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  ]])C*C44>%?%IIST^x^xhhrXsX} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1tt`` L L  88$$  f f""$L%L'')2*2,,./11335q6q88:W;W==?9@9BBDEGGIIKnLnNNPPQPSSU2V2XXZ[]]__agbgddfIgIiik+l+nnp q s~s~uuw`x`zz|B}B$$ttTTĐĒ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2gW2gW2iW2iW2W2W29W29W2;X2p;X2[X2[X2 `X2 `X2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2GY2 GY2Z2p Z2Y[2t Y[2P\2T P\2q"]2q"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v# b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v# b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1v# b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv  hh)Z*Z44>J?JIIS6T6^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1vvhh Z Z  LL>>00""  "#%%'(**,,.q/q113c4c668U9U;;=G>G@@B9C9EEG+H+JJLMOOQRT~T~VVXlYl[[]Z^Z``bFcFeeg0h0jjlmooqrtstsvvxYyY{{}?~?%% {{__ёѓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2kW2 kW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[2;P\2T ;P\2 "]2 "]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jX$ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jX$ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jX$ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  ^^)H*H44>1?1IIST^^hisusu} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qqaa Q Q  AA11 } }""$g%g'')Q*Q,,.;/;113%4%6689;;==?n@nBBDXEXGGIAJALLN)O)QQSTVVXXZm[m]]_U`Ubbd=e=ggi%j%lln o qqssuivixxzQ{Q}}99!! ii b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW25W2 5W2gW2gW2cW2cW2W2W29W29W25X2p5X2[X2[X2 `X2 `X2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[2/P\2T /P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1t b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt  bb)P*P44>>,,  ""$m%m'')[*[,,.I/I113747668%9%;;=>@@BCExExGGIdJdLLNNONQQS8T8VVX"Y"[[] ^ ``bbdgegggiMjMlln1o1qqstvvxxzf{f}}HH** ||^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2W2W2W2W2=W2=W2W2W2qX2pqX2\X2\X2bX2bX2ŢX2`ŢX2%X2%X23X23X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2O\2T O\2!]2!]2q4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jZ& b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jZ& b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1jZ& b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq  \\)F*F44>.?.IIST^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1qq]] I I  55!!  k k""$U%U'')?*?,,.)/)11346688:r;r==?\@\BBDFEFGGI.J.LLNOQQSSUmVmXXZS[S]]_9`9bbdeggijlslsnnpWqWssu;v;xxz{}}uuYỶ̋??%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2mW2mW2cW2cW2W2W2?W2?W25X2p5X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2GY2 GY2Z2p Z2Y[2t Y[2P\2T P\2}"]2}"]2%5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr  bb)R*R44>@?@IIS,T,^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1rrbb R R  BB22""  "#%z%z'')j*j,,.Z/Z113J4J668:9:;;=*>*@@BCEEG H JJLLNlOlQQSXTXVVXDYD[[]0^0``bceeghj~j~llnjojqqsVtVvvxByB{{}.~.||hhޑޓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2׎W2׎W2W2W2CW2CW2W2W2wX2pwX2\X2\X2{bX2{bX2X2`X2X2X2-X2-X2UNY2L UNY25Y2 5Y2 Z2p  Z2G[2t G[2_P\2T _P\2/"]2/"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sa- b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1u} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sa- b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1u} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sa- b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1u } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  [[)I*I44>7?7IIS%T%^^h i s{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm[[ I I  77%% x x""$f%f'')T*T,,.B/B1130406689;;= > @@BBDqEqGGI_J_LLNMOMQQS;T;VVX)Y)[[]^``bce|e|ggijjjllnSoSqqs7t7vvxy{{}}qqUULJlj99 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2/W2 /W2mW2mW2]W2]W2W2W2?W2?W2/X2p/X2[X2[X2`X2`X2X2`X2X2X2-X2-X2UNY2L UNY2/Y2 /Y2Z2p Z2A[2t A[25P\2T 5P\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp  \\)H*H44>2?2IIST^^hisvsv} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1pp\\ H H  44    n n""$Z%Z'')F*F,,.2/21134668 9 ;;==?l@lBBDXEXGGIBJBLLN*O*QQSTVVXXZn[n]]_V`Vbbd>e>ggi&j&llnoqqssujvjxxzR{R}}::"" ~~ff b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2qW2 qW2׎W2׎W2W2W2CW2CW2W2W2wX2pwX2\X2\X2{bX2{bX2X2`X2QX2QX2-X2-X2UNY2L UNY2#Y2 #Y2Z2p Z25[2t 5[2P\2T P\2!]2!]2}4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sc/ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1u} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sc/ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1u} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1sc/ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1u } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  WW)A*A44>+?+IIST^^hhrksk} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmWW A A  ++tt^ ^""$H%H'')2*2,,./11346{6{88:e;e==?O@OBBD9E9GGI#J#LLN O QQSSUlVlXXZV[V]]_@`@bbd*e*ggijllnnpmqmssuQvQxxz5{5}}ooSSŎŐ77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2)W2 )W2[W2[W2]W2]W2W2W2-W2-W2/X2p/X2[X2[X2_X2_X2!X2`!X2X2X2'X2'X2ONY2L ONY25Y2 5Y2 Z2p  Z2G[2t G[2kP\2T kP\2;"]2;"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1p b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  \\)J*J44>6?6IIS T ^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nn\\ J J  88&& y y""$g%g'')U*U,,.C/C1131416689;;= > @@BBDrErGGI^J^LLNHOHQQS2T2VVXY[[]^`{`{bbdeeeggiOjOlln9o9qqs#t#vvx y {{}}llVVˇˉ@@** b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2X2`X2X2X2'X2'X2ONY2L ONY2/Y2 /Y2Z2p Z2A[2t A[2eP\2T eP\25"]25"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|j6 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|j6 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|j6 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  aa)M*M44>7?7IIST^^his{s{} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmaa U U  II==11%%  "#%%'(*{*{,,.g/g113S4S668?9?;;=+>+@@BCEEGHJwJwLLN_O_QQSGTGVVX/Y/[[]^``bbdsesggi[j[llnCoCqqs+t+vvxy{{}}ooWWˇˉ??'' b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2[W2[W2WW2WW2W2W2-W2-W2)X2p)X2[X2[X2_X2_X2X2`X2QX2QX2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2/[2t /[2O\2T O\2!]2!]2w4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1n b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1#  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  YY)C*C44>+?+IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mmYY E E  11  k k""$V%V'')@*@,,.*/*11346688:s;s==?]@]BBDGEGGGI/J/LLNOQQSSUnVnXXZT[T]]_:`:bbd e ggijlylynnp_q_ssuEvExxz+{+}}jj݉݋PPÎÐ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2wW2 wW2W2W2W2W2IW2IW2W2W2}X2p}X2\X2\X2bX2bX2X2`X2KX2KX2'X2'X2ONY2L ONY2Y2 Y2Z2p Z2)[2t )[2O\2T O\2!]2!]2e4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|l8 b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|l8 b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|l8 b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm  ]])E*E44>+?+IIST^^hhrese} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1mm]] M M  ==--   ""$q%q'')Y*Y,,.A/A113)4)6689;;==?m@mBBDUEUGGI;J;LLNOQQSTVuVuXXZY[Y]]_=`=bbd!e!ggijlwlwnnp[q[ssu?v?xxz#{#}}yy]]ωϋAA%% b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2#W2 #W2mW2mW2WW2WW2W2W2?W2?W2)X2p)X2[X2[X2`X2`X2ťX2`ťX2%X2%X2'X2'X2INY2L INY2#Y2 #Y2Z2p Z25[2t 5[2AP\2T AP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1m b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn  ^^)J*J44>2?2IIST^^histst} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1nnbb V V  JJ>>..  "#%|%|'')h*h,,.T/T113@4@668,9,;;=>@@BBDrErGGIZJZLLNBOBQQS*T*VVXY[[]]_n`nbbdVeVggi>j>lln&o&qqstvvxxzj{j}}RRƂƄ::xx b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2}W2 }W2юW2юW2W2W2IW2IW2W2W2}X2p}X2\X2\X2ubX2ubX2ťX2`ťX2%X2%X2'X2'X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2Y4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1v} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1v} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1s? b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1v } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  QQ);*;44>!?!IIST^z^zhhr^s^} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffRR > >  **vv` `""$J%J'')4*4,,./11346}6}88:g;g==?M@MBBD3E3GGIJLLNNPrQrSSUXVXXXZ>[>]]_$`$bbd e g}g}iikclcnnpIqIssu/v/xxz{}}nnRRĉċ66 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2gW2gW2QW2QW2W2W29W29W2#X2p#X2[X2[X2 `X2 `X2]X2`]X2X2X2'X2'X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2A4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1k b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll  XX)@*@44>$?$IIST^z^zhhr\s\} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ll\\ L L  <<,, t t""$\%\'')D*D,,.,/,11346688:p;p==?T@TBBD8E8GGIJLLNOQrQrSSUVVVXXZ:[:]]_`bbdegtgtiikXlXnnp?IISS]l^lhhrLsL} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhTT @ @  ,,ttZ Z""$@%@'')&*&,,. / 11335e6e88:K;K==?-@-BBDEGGIIKbLbNNPDQDSSU&V&XXZ[]y]y__a[b[ddf=g=iiklnnpqsrsruuwTxTzz|6}6hh؋؍HH b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2UW2UW2KW2KW2W2W2'W2'W2X2pX2[X2[X2_X2_X2KX2`KX2X2X2!X2!X2INY2L INY2/Y2 /Y2Z2p Z2A[2t A[2P\2T P\2Y"]2Y"]25]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1j b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1h  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj  \\)N*N44>>?>IIS*T*^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1jj\\ N N  @@22$$  "#%%'')s*s,,.e/e113W4W668I9I;;=;>;@@B-C-EEGHJJL M OOQQSrTrVVX`Y`[[]N^N``b:c:eeg$h$jjlmooqqsgtgvvxMyM{{}3~3ooSSőœ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2W2W2OW2OW2W2W2X2pX2!\X2!\X2bX2bX2EX2`EX2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1|H b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  RR)<*<44>%?%IIS T ^^hhrisi} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeUU E E  55%%q q""$[%['')E*E,,.///11346689;x;x==?b@bBBDLELGGI5J5LLNOQQSTVyVyXXZa[a]]_I`Ibbd1e1ggijllnoququssu]v]xxzE{E}}--ss]] b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2UW2UW2KW2KW2W2W2'W2'W2X2pX2[X2[X2_X2_X2X2`X2X2X2!X2!X2INY2L INY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1h b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh  VV)D*D44>0?0IIST^^hhrjsj} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1hhVV D D  22  s s""$a%a'')O*O,,.=/=113+4+6689;;=>@~@~BBDlElGGIXJXLLNBOBQQS,T,VVXY[[]^`u`ubbd[e[ggiAjAlln%o%qqstvxvxxxzZ{Z}}<<ppRR b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2W2W2W2W2UW2UW2W2W2X2pX2'\X2'\X2bX2bX2X2`X2X2X2!X2!X2INY2L INY2 Y2  Y2Z2p Z2[2t [2O\2T O\2!]2!]2S4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1~J b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee  PP):*:44>"?"IIST^y^yhhr]s]} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1eeQQ = =  ))uu_ _""$I%I'')3*3,,./11346|6|88:f;f==?P@PBBD:E:GGI"J"LLNOQ{Q{SSUaVaXXZG[G]]_-`-bbdeggiikglgnnpKqKssu/v/xxz{}}iiۄۆMM33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2yGW2yGW2W2 W2aW2aW2EW2EW2W2W23W23W2X2pX2[X2[X2`X2`X2ťX2`ťX2%X2%X2X2X2CNY2L CNY2)Y2 )Y2Z2p Z2;[2t ;[2P\2T P\2e"]2e"]2 5]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1g b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff  VV)F*F44>4?4IIS T ^^hiss} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ffVV F F  66&& ~ ~""$n%n'')^*^,,.N/N113>4>668.9.;;=>@@BCEEGGItJtLLN`O`QQSLTLVVX8Y8[[]$^$``bceeggirjrlln`o`qqsLtLvvx8y8{{}$~$rr^^ԑԓ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2ŎW2ŎW2W2W2UW2UW2W2W2X2pX2'\X2'\X2ibX2ibX2X2`X2X2X2X2X2CNY2L CNY2#Y2 #Y2Z2p Z25[2t 5[2MP\2T MP\2"]2"]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1u} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1u} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1u } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  OO)=*=44>+?+IIST^^hisysy} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaOO = =  ++~~l l""$Z%Z'')H*H,,.6/6113$4$6689;;=>@w@wBBDeEeGGISJSLLNAOAQQS/T/VVXY[[] ^ ``bbdpepggi^j^llnQoQqqs5t5vvxy{{}}ooSSŇʼn77 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2sGW2sGW2W2 W2aW2aW2EW2EW2W2W23W23W2X2pX2[X2[X2`X2`X2X2`X2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2)[2t )[2P\2T P\2!]2!]24]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1e b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd  PP)<*<44>&?&IIST^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1ddPP < <  ((vvb b""$N%N''):*:,,.&/&11346688:t;t==?`@`BBDLELGGI6J6LLNOQQSTVzVzXXZb[b]]_J`Jbbd2e2ggijllnoqzqzssubvbxxzJ{J}}22vv^^ b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2GW2GW2W2 W2ŎW2ŎW2W2W2UW2UW2W2W2X2pX2'\X2'\X2ibX2ibX2qX2`qX2X2X2X2X2CNY2L CNY2Y2 Y2Z2p Z2#[2t #[2O\2T O\2!]2!]2k4]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1u} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1u} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1S b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1u } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  KK)5*544>?IIS T ^^hhrgsg} b)woperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaKK 5 5    ~~hhR R""$<%<'')&*&,,./11335o6o88:Y;Y==?C@CBBD-E-GGIJLLNOQvQvSSU`V`XXZJ[J]]_4`4bbdeggijl}l}nnpiqissuMvMxxz1{1}}kk݉݋OO33 b)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L2W2mGW2mGW2 W2  W2OW2OW2?W2?W2W2W2!W2!W2X2pX2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2#Y2 #Y2KZ2p KZ2)[2t )[2YP\2T YP\2O ]2O ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1d b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb  PP)@*@44>,?,IIST^^h i ss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1bbPP > >  ,,m m""$]%]'')K*K,,.9/9113'4'6689;;=>@z@zBBDhEhGGITJTLLN>O>QQS(T(VVXY[[]]_q`qbbd[e[ggiEjElln;o;qqs%t%vvxy{{}}nnXX͇͉BB,, b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2[W2[W2W2W2X2pX2-\X2-\X2bX2bX2X2`X2WX2WX2X2X2=NY2L =NY2)Y2 )Y2QZ2p QZ2/[2t /[2_P\2T _P\2U ]2U ]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1Z b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1A=  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  UU)K*K44>5?5IIST^^hiss} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaUU I I  ==11%%  " # %%'(*y*y,,.e/e113Q4Q668=9=;;=)>)@@BCEEGHJuJuLLN]O]QQSETEVVX-Y-[[]^``bbdqeqggiYjYllnMoMqqs5t5vvxy{{}~yyaaՇՉII11 b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2mGW2mGW2 W2  W2OW2OW29W29W2W2W2!W2!W2 X2p X2[X2[X2_X2_X2X2`X2X2X2X2X2=NY2L =NY2Y2 Y29Z2p 9Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1b  b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}"}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  MM);*;44>#?#IIS T ^^hhrnsn} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaMM 9 9  %%ss_ _""$N%N'')8*8,,."/"113 4 6688:k;k==?U@UBBD?E?GGI'J'LLN O QQSSUfVfXXZL[L]]_2`2bbdeggiikqlqnnpcqcssuIvIxxz/{/}}nnTTǎǐ:: b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L2W2GW2GW2W2 W2W2W2W2W2[W2[W2W2W2X2pX2-\X2-\X2bX2bX2X2`X2X2X2X2X2=NY2L =NY2Y2 Y29Z2p 9Z2[2t [2O\2T O\2]2]22]2} bJ7U'operation.0exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1} bJ7U'operation.1exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1} bJ7U'operation.2exchange :!2!!b%reasonOperatorInput { input: 0 }value0invocations1\ b-woperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=yz}*}z2z*breasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 } bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  b-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5z"z"breasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements504input0Elements504invocations1  b*7Ioperation.4sync :2breasonExchangeBarriervalue8invocations1  b*7Ioperation.5sync :2breasonExchangeBarriervalue9invocations1 } bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa  QQ)A*A44>'?'IIS T ^^hhrmsm} b)woperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=yz}*}z"z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b)goperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5aze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 bJ7Moperation.6exchange :z2breasonOperatorInputsvalue8invocations1aaQQ A A  11!! y y""$m%m'')U*U,,.=/=113%4%668 9 ;;==?i@iBBDQEQGGI7J7LLNOQQSSUqVqXXZU[U]]_9`9bbdeggijlslsnnpcqcssuGvGxxz+{+}}ee׉׋II-- b)woperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=yz}"}z*z2breasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%woperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z"z"z*breasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1